钱学森实验室:十年求索,追求空间科学与技术进步永无止境

发布时间:2022-7-04 | 杂志分类:其他
免费制作
更多内容

钱学森实验室:十年求索,追求空间科学与技术进步永无止境

Lianchong Zhang Aerospace Information Research Institute, Chinese Academy of Sciences, China Xiaopeng Xue Central South University, China Ruoxi Tian Beijing Institute of Technology Press, China Ruixue Bai Beijing SpaceD Aerospace Application & Science Education Co.Ltd., China Yudong Feng Lanzhou Institute of Physics, CAST, China Yuqing Xiong Lanzhou Institute of Physics, CAST, China Guangping Zhao Lanzhou Institute of Physics, CAST, China Publishing Editors Email: space@science-bitpjo... [收起]
[展开]
钱学森实验室:十年求索,追求空间科学与技术进步永无止境
粉丝: {{bookData.followerCount}}
文本内容
第0页
第1页
第2页
第3页
// / // / / :
第4页
Yiming Lin
第5页
Yiming Lin : Hongtai Zhang China Satellite Network Group Co., Ltd., China
第6页
University of Luxembourg, Luxembourg
第7页
Lianchong Zhang Aerospace Information Research Institute, Chinese Academy of Sciences, China Xiaopeng Xue Central South University, China Ruoxi Tian Beijing Institute of Technology Press, China Ruixue Bai Beijing SpaceD Aerospace Application & Science Education Co.Ltd., China Yudong Feng Lanzhou Institute of Physics, CAST, China Yuqing Xiong Lanzhou Institute of Physics, CAST, China Guangping Zhao Lanzhou Institute of Physics, CAST, China Publishing Editors Email: space@science-bitpjournal.org.cn Tel: 010-68948375
第8页
第9页
Space: Science & Technology ( Special Issue ) CONTENTS (1) (3) (13) (19) (31) Editorial Decadal Advances and Review: Pursuing Space Science and Technology Innovation A Bridge Neural Network-Based Optical-SAR Image Joint Intelligent Interpretation Framework Meiyu Huang, Yao Xu, Lixin Qian, Weili Shi, Yaqin Zhang, Wei Bao, Nan Wang, Xuejiao Liu, and Xueshuang Xiang Developing Prototype Simulants for Surface Materials and Morphology of Near Earth Asteroid 2016 HO3 Xiaojing Zhang, Yuechen Luo, Yuan Xiao, Deyun Liu, Fan Guo, and Qian Guo Performance Evaluation Indicators of Space Dynamic Networks under Broadcast Mechanism Zipeng Ye and Qingrui Zhou Mission Design of an Aperture-Synthetic Interferometer System for Space-Based Exoplanet Exploration Feida Jia, Xiangyu Li, Zhuoxi Huo, and Dong Qiao Decentralized Distributed Deep Learning with Low-Bandwidth Consumption for Smart Constellations Qingliang Meng, Meiyu Huang, Yao Xu, Naijin Liu, and Xueshuang Xiang A Review on the Development of Spaceborne Membrane Antennas Ming-Jian Li, Meng Li, Yu-Fei Liu, Xin-Yu Geng, and Yuan-Yuan Li Unsupervised Spectrum Anomaly Detection Method for Unauthorized Bands Yu Tian, Haihua Liao, Jing Xu, Ya Wang, Shuai Yuan, and Naijin Liu (43) (53) (69)
第10页
第11页
1 Editorial Decadal Advances and Review: Pursuing Space Science and Technology Innovation This special issue is dedicated to Qian Xuesen Laboratory of Space Technology (hereinafter referred to as Qian Lab) on its 10th anniversary. Qian Lab was established in December 2011, in honor of the 100 years birth anniversary of the worldrenowned scientist, Dr. Qian Xuesen. As the “Special Innovative Zone” of China Academy of Space Technology, Qian Lab aims to bloom into a world-class laboratory of space science and technology. Qian Lab remains in the frontier of the transition of the Chinese space industry in the past decade. In the faith to cater the major national needs and the shared welfare of all mankind, Qian Lab has concentrating on: space strategic planning, space system and metasystem demonstrations, engineering sciences, and frontier technologies. Six academicians from the Chinese Academy of Sciences and Chinese Academy of Engineering lead the research, together with over 220 full-time researchers and about 50 students. Facing flourish opportunities of the future space systems and missions, Qian Lab considers itself the bridge between space industry and research in China and abroad. It works vigorously to strengthen close cooperation with partners and forms close bonds. Qian Lab has contributed greatly to the metasystem demonstration for the civil space and the future National Science and Technology Major Project of China. Articles within one special issue cannot cover all the research fields involved in Qian Lab. However, this special issue presents what Qian Lab has been always pursuing, devoting to the progression of space science and technology, and of course, missions. We elaborately select six research articles that cover the following specific topics: space interferometer mission design, Near-Earth asteroid surface science, artificial intelligence for space-borne remote sensing, networks and smart constellation, deployable spaceborne membrane antenna. These articles reflect the joint efforts and multidisciplinary research approaches in Qian Lab. “How did Nine Spheres divide and join up side by side?”, which is asked by ancient Chinese poet Qu Yuan in his masterpiece Heavenly Questions (Tian Wen) thousands of years ago. The question is also raised by the crew of Qian Lab, who strive for a better understanding of our place in the universe, enable future space explorations, and benefit life on Earth. A decade remains just another starting point, and Qian Lab’s people will remain dedicated and committed to the development of the Chinese space industry and the well-being of the humanity. We would like to further express our gratitude to Dr. Ye Peijian and Dr. Bao Weimin, the Editor-in-Chief and the Guest Editor of Space: Science and Technology for their cordial and professional support to this special issue. Our gratitude also goes to the whole editorial team for their professionalism and enthusiasm.
第12页
2 Research Article A Bridge Neural Network-Based Optical-SAR Image Joint Intelligent Interpretation Framework Meiyu Huang , Yao Xu, Lixin Qian , Weili Shi, Yaqin Zhang, Wei Bao , Nan Wang, Xuejiao Liu, and Xueshuang Xiang Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing, China Correspondence should be addressed to Xueshuang Xiang; xiangxueshuang@qxslab.cn Received 4 August 2021; Accepted 23 September 2021; Published 12 October 2021 Copyright © 2021 Meiyu Huang et al. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). The current interpretation technology of remote sensing images is mainly focused on single-modal data, which cannot fully utilize the complementary and correlated information of multimodal data with heterogeneous characteristics, especially for synthetic aperture radar (SAR) data and optical imagery. To solve this problem, we propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, optimizing the feature correlation between optical and SAR images through optical-SAR matching tasks. It adopts BNN to effectively improve the capability of common feature extraction of optical and SAR images and thus improving the accuracy and application scenarios of specific intelligent interpretation tasks for optical-SAR/SAR/optical images. Specifically, BNN projects optical and SAR images into a common feature space and mines their correlation through pair matching. Further, to deeply exploit the correlation between optical and SAR images and ensure the great representation learning ability of BNN, we build the QXS-SAROPT dataset containing 20,000 pairs of perfectly aligned optical-SAR image patches with diverse scenes of high resolutions. Experimental results on optical-to-SAR crossmodal object detection demonstrate the effectiveness and superiority of our framework. In particular, based on the QXSSAROPT dataset, our framework can achieve up to 96% high accuracy on four benchmark SAR ship detection datasets. 1. Introduction With the rapid development of deep learning, remarkable breakthroughs have been made in deep learning-based land use segmentation, scene classification, object detection, and recognition in thefield of remote sensing in the past decade [1–4]. This is mainly due to the powerful feature extraction and representation ability of deep neural networks [5–8], which can well map the remote sensing observations into the desired geographical knowledge. However, the current mainstream interpretation technology for remote sensing images is still mainly focused on single-modal data and cannot make full use of the complementary and correlated information of multimodal data from different sensors with heterogeneous characteristics, resulting in insufficient intelligent interpretation capabilities and limited application scenarios. For example, optical imaging is easily restricted by illumination and weather conditions, based on which accurate interpretation cannot be obtained at night or under complex weather with clouds, fog, and so on. Compared with optical imaging, synthetic aperture radar (SAR) imaging can achieve full-time and all-weather earth observations. However, due to the lack of texture features, it is difficult for SAR images to be interpreted even by well-trained experts. Therefore, gathering sufficient amounts of training SAR data with diverse scenes and accurate labeling is a challenging problem, which heavily affects the deep research and application of SAR images based on intelligent interpretation. To address the above issues, multimodal data fusion [9–12] becomes one of the most promising application directions of deep learning in remote sensing, especially the combined utilization of SAR and optical data because these data modalities are completely different from each other in terms of geometric and radiometric appearance [13–17]. However, the existing optical-SAR fusion techniques mainly concentrate on the matching problem. The proposed optical-SAR image matching methods can be divided into three types: signal-based, hand-crafted feature-based, and AAAS Space: Science & Technology Volume 2021, Article ID 9841456, 10 pages https://doi.org/10.34133/2021/9841456
第13页
3 Research Article A Bridge Neural Network-Based Optical-SAR Image Joint Intelligent Interpretation Framework Meiyu Huang , Yao Xu, Lixin Qian , Weili Shi, Yaqin Zhang, Wei Bao , Nan Wang, Xuejiao Liu, and Xueshuang Xiang Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing, China Correspondence should be addressed to Xueshuang Xiang; xiangxueshuang@qxslab.cn Received 4 August 2021; Accepted 23 September 2021; Published 12 October 2021 Copyright © 2021 Meiyu Huang et al. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). The current interpretation technology of remote sensing images is mainly focused on single-modal data, which cannot fully utilize the complementary and correlated information of multimodal data with heterogeneous characteristics, especially for synthetic aperture radar (SAR) data and optical imagery. To solve this problem, we propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, optimizing the feature correlation between optical and SAR images through optical-SAR matching tasks. It adopts BNN to effectively improve the capability of common feature extraction of optical and SAR images and thus improving the accuracy and application scenarios of specific intelligent interpretation tasks for optical-SAR/SAR/optical images. Specifically, BNN projects optical and SAR images into a common feature space and mines their correlation through pair matching. Further, to deeply exploit the correlation between optical and SAR images and ensure the great representation learning ability of BNN, we build the QXS-SAROPT dataset containing 20,000 pairs of perfectly aligned optical-SAR image patches with diverse scenes of high resolutions. Experimental results on optical-to-SAR crossmodal object detection demonstrate the effectiveness and superiority of our framework. In particular, based on the QXSSAROPT dataset, our framework can achieve up to 96% high accuracy on four benchmark SAR ship detection datasets. 1. Introduction With the rapid development of deep learning, remarkable breakthroughs have been made in deep learning-based land use segmentation, scene classification, object detection, and recognition in thefield of remote sensing in the past decade [1–4]. This is mainly due to the powerful feature extraction and representation ability of deep neural networks [5–8], which can well map the remote sensing observations into the desired geographical knowledge. However, the current mainstream interpretation technology for remote sensing images is still mainly focused on single-modal data and cannot make full use of the complementary and correlated information of multimodal data from different sensors with heterogeneous characteristics, resulting in insufficient intelligent interpretation capabilities and limited application scenarios. For example, optical imaging is easily restricted by illumination and weather conditions, based on which accurate interpretation cannot be obtained at night or under complex weather with clouds, fog, and so on. Compared with optical imaging, synthetic aperture radar (SAR) imaging can achieve full-time and all-weather earth observations. However, due to the lack of texture features, it is difficult for SAR images to be interpreted even by well-trained experts. Therefore, gathering sufficient amounts of training SAR data with diverse scenes and accurate labeling is a challenging problem, which heavily affects the deep research and application of SAR images based on intelligent interpretation. To address the above issues, multimodal data fusion [9–12] becomes one of the most promising application directions of deep learning in remote sensing, especially the combined utilization of SAR and optical data because these data modalities are completely different from each other in terms of geometric and radiometric appearance [13–17]. However, the existing optical-SAR fusion techniques mainly concentrate on the matching problem. The proposed optical-SAR image matching methods can be divided into three types: signal-based, hand-crafted feature-based, and AAAS Space: Science & Technology Volume 2021, Article ID 9841456, 10 pages https://doi.org/10.34133/2021/9841456 A Bridge Neural Network-Based Optical-SAR Image Joint Intelligent Interpretation Framework Meiyu Huang, Yao Xu, Lixin Qian, Weili Shi, Yaqin Zhang, Wei Bao, Nan Wang, Xuejiao Liu, and Xueshuang Xiang Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing, China Correspondence should be addressed to Xueshuang Xiang; xiangxueshuang@qxslab.cn Abstract: The current interpretation technology of remote sensing images is mainly focused on single-modal data, which cannot fully utilize the complementary and correlated information of multimodal data with heterogeneous characteristics, especially for synthetic aperture radar (SAR) data and optical imagery. To solve this problem, we propose a bridge neural network-(BNN-) based optical-SAR image joint intelligent interpretation framework, optimizing the feature correlation between optical and SAR images through opticalSAR matching tasks. It adopts BNN to effectively improve the capability of common feature extraction of optical and SAR images and thus improving the accuracy and application scenarios of specific intelligent interpretation tasks for optical-SAR/SAR/optical images. Specifically, BNN projects optical and SAR images into a common feature space and mines their correlation through pair matching. Further, to deeply exploit the correlation between optical and SAR images and ensure the great representation learning ability of BNN, we build the QXS-SAROPT dataset containing 20,000 pairs of perfectly aligned optical-SAR image patches with diverse scenes of high resolutions. Experimental results on optical-to-SAR crossmodal object detection demonstrate the effectiveness and superiority of our framework. In particular, based on the QXSSAROPT dataset, our framework can achieve up to 96% high accuracy on four benchmark SAR ship detection datasets.
第14页
4 deep learning-based approaches. For the signal-based similarity measures, crosscorrelation (CC) [18] and mutual information (MI) [19, 20] have been widely used to the optical-SAR matching tasks. Since MI is an intensity-based statistical measure and has good adaptability to geometric and radiometric changes, it is extensively outperformed CC in optical-SAR image matching. Nevertheless, signal-based approaches that do not contain any local structure information are not robust and accurate enough for matching multisensor images. Feature-based methods commonly utilize invariant key points and feature descriptors. The reason why it comes to a better result may stem from the feature descriptors, which are less sensitive to the geometric and radiometric changes. Many traditional handcrafted methods have been proposed for optical-SAR image matching, such as SIFT [21], SAR-SIFT [22], and HOPC [23, 24]. However, considering highly divergence between SAR and optical images and the computing ability, the hand-crafted feature-based matching approaches are quite limited to get a further step. Because of the powerful feature extraction and representation learning ability, exploiting convolution neural networks to extract the deep features has achieved a high matching accuracy. The mainstream architecture for optical-SAR image matching is Siamese network [25–28], which is composed of two identical convolutional streams. The dual network is used to extract deep characteristic information of input image pairs; therefore, the deep features in the same space can be measured under the same metric. However, the Siamese network can be only applied to the optical-SAR image matching problem; yet, no subsequent optical-SAR images joint interpretation work. Based on the analysis above, we innovatively propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, which utilizes BNN to enhance the general feature embedding of optical and SAR images to improve the accuracy and application scenarios of specific optical-SAR images joint intelligent interpretation tasks. Completely different from the Siamese network, BNN contains two independent feature extraction networks and projects the optical and SAR images into a subspace to learn the desired common representation where features can be measured with Euclidean distance. The proposed framework is shown in Figure 1. BNN is trained on an optical-SAR image matching dataset to learn the common representation from optical and SAR images so that the BNN model can be transferred to the feature extraction module forfine-tuning the interpretation model on optical-SAR/SAR/optical image interpretation datasets. Further, to verify the effectiveness and the superiority of our proposed framework and promote the development of research in optical-SAR image fusion based on deep learning, it is very important to obtain datasets of a large number of perfectly aligned optical-SAR images. Besides, considering the existing optical-SAR image matching dataset either lacks scene diversity due to the huge difficulty in pixel-level matching between optical and SAR images [29], or has a low resolution limited by the remote sensing satellites [14], or covers only a single area [30], which cannot fully exploit the relevance of optical and SAR images, we publish the QXS-SAROPT dataset, which contains 20,000 optical-SAR patch pairs from multiple scenes of a high resolution. Specifically, the SAR images are collected from the Gaofen-3 satellite [31], and the corresponding optical images are from the Google Earth [32]. These images spread across landmasses of San Diego, Shanghai, and Qingdao. The QXS-SAROPT dataset under open access license CCBY is publicly available at https://github.com/yaoxu008/QXS-SAROPT. On this basis, we conduct experiments on the optical-toSAR crossmodal object detection to demonstrate the effectiveness and superiority of our framework. In particular, based on the QXS-SAROPT dataset, our framework can achieve up to 96% high accuracy on four benchmark SAR ship detection datasets. The contributions of this paper can be summarized as follows: (i) We propose a BNN-based optical-SAR image joint intelligent interpretation framework, which can effectively improve the generic feature extraction capability of optical and SAR images, and thus improving the accuracy and the application scenarios of specific intelligent interpretation tasks for optical-SAR/SAR/optical images (ii) We publish the optical-SAR matching dataset: QXSSAROPT, which contains 20,000 optical-SAR image pairs from multiple scenes of a high resolution of 1 meter to support the joint interpretation of optical and SAR images (iii) The BNN-based optical-SAR image joint intelligent interpretation framework is applied to SAR ship detection and achieves high accuracy on four SAR ship detection benchmark datasets Optical-SAR/SAR/optical image interpretation dataset Interpretation results Feature extraction module Interpretation module BNN Optical-SAR image matching dataset Common representations Interpretation model Transfer Figure 1: The BNN-based optical-SAR image joint intelligent interpretation framework. 2 Space: Science & Technology 2. Methodology In this section, the details of the bridge neural network (BNN) and the proposed BNN-based joint interpretation framework are introduced. 2.1. Bridge Neural Network. The bridge neural network (BNN) proposed in [33] is adopted to learn the common representations of optical and SAR images on the opticalSAR image matching tasks, as shown in Figure 2. Given a SAR-optical image matching dataset fXs, Xog, where Xs = fxi sg N i=1 is the set of SAR images and Xo = fxi og N i=1 is the set of corresponding optical images, here, we consider samples from Sp = fxi s, xi og, whose image pairs from the same region, as positive samples and samples from Sn = fxi s, xj og ði≠ jÞ as negative samples. Different from Siamese network, BNN contains two separate feature extraction networks: SAR network f s ðxs ;θsÞ and optical network f oðxo ; θoÞ with parameters ðθs,θoÞ, respectively, extracting features from SAR and optical images ðxs, xoÞ. To decrease the feature dimension, following the feature extraction backbone, we employ a 1×1 convolution layer and a 4×4 averagepooling layer to the feature map. Finally, a linear layer with sigmoid activation function is followed to project the feature map into the n-dimension common feature representations: zs = f s ðxs,θsÞ, zo = f oðxo,θoÞ. Then, the BNN outputs the Euclidean distance between zs and zo to measure the relevance of the input SAR-optical image pairs, which is described as f xs, xo ;θs,θo ð Þ = 1 ffiffiffi n p f s xs ;θs ð Þ− f o xo ;θo k k ð Þ ð Þ 2, ð1Þ where n is the dimension of zs, zo. The Euclidean distance indicates whether the input data pairs fxs, xog have a potential relation. And the closer distance between them, the more relevant they are. Specifically, the distance between positive samples tends to 0 while the distance between negative samples is close to 1. Therefore, the loss on positive samples and negative samples is set as follows: lp Sp ;θs,θo � � = 1 Sp � �� � 〠 xs,xo ð Þ∈Sp f xs, xo ;θs,θo ð Þ ð Þ− 0 2 , ð2Þ ln Sn ;θs,θo ð Þ = 1 Sn j j〠 xs,xo ð Þ∈Sn f xs, xo ;θs,θo ð Þ ð Þ− 1 2 : ð3Þ Hence, the problem of learning the common representations of SAR-optical images is taken as a binary classification problem. The overall loss of BNN can be written as lBNN Sp, Sn ;θs,θo � � = lp Sp ;θs,θo � � +α · ln Sn ;θs,θo ð Þ 1 +α , ð4Þ whereα is a hyperparameter to balance the weights of positive loss and negative loss. Then, the best weights ðθ∗ s ,θ∗ o Þ can be obtained via a optimization issue: θ∗ s ,θ∗ o ð Þ = arg minθs,θo lBNN Sp, Sn ;θs,θo � �: ð5Þ 2.2. Optical-SAR Image Joint Intelligent Interpretation. Since BNN projects optical-SAR image patches into a common feature subspace, the model can well mine the correlation between optical and SAR images and thus improving the feature learning ability of optical-SAR images. Based on BNN, we propose the optical-SAR image joint interpretation framework, which can jointly utilize optical and SAR features enhanced by BNN and improve the performance of specific interpretation tasks of optical-SAR/SAR/optical images. As depicted in Figure 3, we present two different usage scenarios of the proposed BNN-based optical-SAR image joint intelligent interpretation framework. As shown in Figure 3(a), for applications of optical and SAR fusion intelligent interpretation, such as object detection, classification, and segmentation, our framework can learn the common representations from SAR and optical images and enhance their feature learning ability for better interpretation performance under all-weather in full-time. As for applications of crossmodal intelligent interpretation, see Figure 3(b), our frameworkfirst utilizes BNN to project optical and SAR images into a common feature space and mines their complementary and correlated information through optical-SAR image matching. Then, the feature 0 Negative sample: unmatched pairs Optical CNN SAR CNN 1 Positive sample: matched pairs Figure 2: Illustration of BNN structure on the SAR-optical matching task. BNN contains two feature extraction networks: optical CNN and SAR CNN, to project the images into a common representation space. The Euclidean distance of the representations of SAR and optical images is regressed to 0 for positive samples and to 1 for negative samples, respectively. Space: Science & Technology 3
第15页
5 deep learning-based approaches. For the signal-based similarity measures, crosscorrelation (CC) [18] and mutual information (MI) [19, 20] have been widely used to the optical-SAR matching tasks. Since MI is an intensity-based statistical measure and has good adaptability to geometric and radiometric changes, it is extensively outperformed CC in optical-SAR image matching. Nevertheless, signal-based approaches that do not contain any local structure information are not robust and accurate enough for matching multisensor images. Feature-based methods commonly utilize invariant key points and feature descriptors. The reason why it comes to a better result may stem from the feature descriptors, which are less sensitive to the geometric and radiometric changes. Many traditional handcrafted methods have been proposed for optical-SAR image matching, such as SIFT [21], SAR-SIFT [22], and HOPC [23, 24]. However, considering highly divergence between SAR and optical images and the computing ability, the hand-crafted feature-based matching approaches are quite limited to get a further step. Because of the powerful feature extraction and representation learning ability, exploiting convolution neural networks to extract the deep features has achieved a high matching accuracy. The mainstream architecture for optical-SAR image matching is Siamese network [25–28], which is composed of two identical convolutional streams. The dual network is used to extract deep characteristic information of input image pairs; therefore, the deep features in the same space can be measured under the same metric. However, the Siamese network can be only applied to the optical-SAR image matching problem; yet, no subsequent optical-SAR images joint interpretation work. Based on the analysis above, we innovatively propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, which utilizes BNN to enhance the general feature embedding of optical and SAR images to improve the accuracy and application scenarios of specific optical-SAR images joint intelligent interpretation tasks. Completely different from the Siamese network, BNN contains two independent feature extraction networks and projects the optical and SAR images into a subspace to learn the desired common representation where features can be measured with Euclidean distance. The proposed framework is shown in Figure 1. BNN is trained on an optical-SAR image matching dataset to learn the common representation from optical and SAR images so that the BNN model can be transferred to the feature extraction module forfine-tuning the interpretation model on optical-SAR/SAR/optical image interpretation datasets. Further, to verify the effectiveness and the superiority of our proposed framework and promote the development of research in optical-SAR image fusion based on deep learning, it is very important to obtain datasets of a large number of perfectly aligned optical-SAR images. Besides, considering the existing optical-SAR image matching dataset either lacks scene diversity due to the huge difficulty in pixel-level matching between optical and SAR images [29], or has a low resolution limited by the remote sensing satellites [14], or covers only a single area [30], which cannot fully exploit the relevance of optical and SAR images, we publish the QXS-SAROPT dataset, which contains 20,000 optical-SAR patch pairs from multiple scenes of a high resolution. Specifically, the SAR images are collected from the Gaofen-3 satellite [31], and the corresponding optical images are from the Google Earth [32]. These images spread across landmasses of San Diego, Shanghai, and Qingdao. The QXS-SAROPT dataset under open access license CCBY is publicly available at https://github.com/yaoxu008/QXS-SAROPT. On this basis, we conduct experiments on the optical-toSAR crossmodal object detection to demonstrate the effectiveness and superiority of our framework. In particular, based on the QXS-SAROPT dataset, our framework can achieve up to 96% high accuracy on four benchmark SAR ship detection datasets. The contributions of this paper can be summarized as follows: (i) We propose a BNN-based optical-SAR image joint intelligent interpretation framework, which can effectively improve the generic feature extraction capability of optical and SAR images, and thus improving the accuracy and the application scenarios of specific intelligent interpretation tasks for optical-SAR/SAR/optical images (ii) We publish the optical-SAR matching dataset: QXSSAROPT, which contains 20,000 optical-SAR image pairs from multiple scenes of a high resolution of 1 meter to support the joint interpretation of optical and SAR images (iii) The BNN-based optical-SAR image joint intelligent interpretation framework is applied to SAR ship detection and achieves high accuracy on four SAR ship detection benchmark datasets Optical-SAR/SAR/optical image interpretation dataset Interpretation results Feature extraction module Interpretation module BNN Optical-SAR image matching dataset Common representations Interpretation model Transfer Figure 1: The BNN-based optical-SAR image joint intelligent interpretation framework. 2 Space: Science & Technology 2. Methodology In this section, the details of the bridge neural network (BNN) and the proposed BNN-based joint interpretation framework are introduced. 2.1. Bridge Neural Network. The bridge neural network (BNN) proposed in [33] is adopted to learn the common representations of optical and SAR images on the opticalSAR image matching tasks, as shown in Figure 2. Given a SAR-optical image matching dataset fXs, Xog, where Xs = fxi sg N i=1 is the set of SAR images and Xo = fxi og N i=1 is the set of corresponding optical images, here, we consider samples from Sp = fxi s, xi og, whose image pairs from the same region, as positive samples and samples from Sn = fxi s, xj og ði≠ jÞ as negative samples. Different from Siamese network, BNN contains two separate feature extraction networks: SAR network f s ðxs ;θsÞ and optical network f oðxo ; θoÞ with parameters ðθs,θoÞ, respectively, extracting features from SAR and optical images ðxs, xoÞ. To decrease the feature dimension, following the feature extraction backbone, we employ a 1×1 convolution layer and a 4×4 averagepooling layer to the feature map. Finally, a linear layer with sigmoid activation function is followed to project the feature map into the n-dimension common feature representations: zs = f sðxs,θsÞ, zo = f oðxo,θoÞ. Then, the BNN outputs the Euclidean distance between zs and zo to measure the relevance of the input SAR-optical image pairs, which is described as f xs, xo ;θs,θo ð Þ = 1 ffiffiffi n p f s xs ;θs ð Þ− f o xo ;θo k k ð Þ ð Þ 2, ð1Þ where n is the dimension of zs, zo. The Euclidean distance indicates whether the input data pairs fxs, xog have a potential relation. And the closer distance between them, the more relevant they are. Specifically, the distance between positive samples tends to 0 while the distance between negative samples is close to 1. Therefore, the loss on positive samples and negative samples is set as follows: lp Sp ;θs,θo � � = 1 Sp � �� � 〠 xs,xo ð Þ∈Sp f xs, xo ;θs,θo ð Þ ð Þ− 0 2 , ð2Þ ln Sn ;θs,θo ð Þ = 1 Sn j j〠 xs,xo ð Þ∈Sn f xs, xo ;θs,θo ð Þ ð Þ− 1 2 : ð3Þ Hence, the problem of learning the common representations of SAR-optical images is taken as a binary classification problem. The overall loss of BNN can be written as lBNN Sp, Sn ;θs,θo � � = lp Sp ;θs,θo � � +α · ln Sn ;θs,θo ð Þ 1 +α , ð4Þ whereα is a hyperparameter to balance the weights of positive loss and negative loss. Then, the best weights ðθ∗ s ,θ∗ o Þ can be obtained via a optimization issue: θ∗ s ,θ∗ o ð Þ = arg minθs,θo lBNN Sp, Sn ;θs,θo � �: ð5Þ 2.2. Optical-SAR Image Joint Intelligent Interpretation. Since BNN projects optical-SAR image patches into a common feature subspace, the model can well mine the correlation between optical and SAR images and thus improving the feature learning ability of optical-SAR images. Based on BNN, we propose the optical-SAR image joint interpretation framework, which can jointly utilize optical and SAR features enhanced by BNN and improve the performance of specific interpretation tasks of optical-SAR/SAR/optical images. As depicted in Figure 3, we present two different usage scenarios of the proposed BNN-based optical-SAR image joint intelligent interpretation framework. As shown in Figure 3(a), for applications of optical and SAR fusion intelligent interpretation, such as object detection, classification, and segmentation, our framework can learn the common representations from SAR and optical images and enhance their feature learning ability for better interpretation performance under all-weather in full-time. As for applications of crossmodal intelligent interpretation, see Figure 3(b), our frameworkfirst utilizes BNN to project optical and SAR images into a common feature space and mines their complementary and correlated information through optical-SAR image matching. Then, the feature 0 Negative sample: unmatched pairs Optical CNN SAR CNN 1 Positive sample: matched pairs Figure 2: Illustration of BNN structure on the SAR-optical matching task. BNN contains two feature extraction networks: optical CNN and SAR CNN, to project the images into a common representation space. The Euclidean distance of the representations of SAR and optical images is regressed to 0 for positive samples and to 1 for negative samples, respectively. Space: Science & Technology 3
第16页
6 extraction module of BNN for the image modality to be interpreted is used as the pretained model for feature embedding of the specific crossmodal intelligent interpretation task. In this way, plentiful complementary features transferred from images of the other modality during the learning of common feature space can be used to enhance the feature embeddings to be interpreted. Benefited from the enhanced feature embeddings, our framework can effectively improve the interpretation performance of crossmodal SAR/optical images. Take SAR ship detection as an example. SAR ship detection in complex scenes is a great challenging task. And CNN-based SAR ship detection methods have drawn considerable attention because of the powerful feature embedding ability. Due to the scarce labeled SAR images, the pretraining technique is adopted to support these CNNbased SAR ship detectors. As SAR completely different from optical images, directly leveraging ImageNet [34] pretraining is hardly to obtain a good ship detector. However, our proposed framework can transfer rich texture features from optical images to SAR images to obtain a specific feature extraction model with better SAR feature embedding capabilities. Specifically, our proposed framework resorts to a SAR feature embedding operator from common representation learning based on the optical-SAR image matching task using BNN. 3. QXS-SAROPT Dataset To fully exploit the relevance of optical and SAR images and verify the effectiveness of our proposed framework, a large perfectly aligned optical-SAR image dataset with diverse scenes of a high resolution is in need. Considering the existing optical-SAR image matching dataset either lack scene diversity or has a low resolution, we have published the QXS-SAROPT dataset, which contains 20,000 pairs of SAR and optical image patches with a high resolution of 1 m extracted from multiple Gaofen-3 and Google Earth [32] scenes. As far as we know, QXS-SAROPT is thefirst dataset to provide high-resolution 1m × 1m coregistered SAR and optical satellite image patches covering over three big port cities in the world: San Diego, Shanghai, and Qingdao. The coverage of these images is shown in Figure 4. Algorithm 1 shows the procedure for the QXS-SAROPT dataset construction. Finally, 20,000 high-quality image patch pairs are preserved in our dataset, some of which are shown in Figure 5 for examples. 4. Experiments To verify the effectiveness and superiority of our BNN-based optical-SAR image joint intelligent interpretation framework, Object detection SAR image Optical image (a) (b) Common feature space SAR image Optical image SAR image Common feature space Figure 3: Two usage scenarios of the proposed BNN-based optical-SAR image joint intelligent interpretation framework. (a) Example illustration of the proposed framework applied to optical and SAR fusion intelligent interpretation. (b) Example illustration of the proposed framework applied to optical to SAR cross-modal intelligent interpretation. (a) (b) (c) Figure 4: Coverage of the SAR images of Gaofen-3 for constructing the QXS-SAROPT dataset. (a) SanDiego. (b) Shanghai. (c) Qingdao. 4 Space: Science & Technology the framework is applied to one typical optical-to-SAR crossmodal object detection task. We conduct optical-to-SAR crossmodal object detection tasks on four benchmark SAR ship detection datasets. We select two representative CNN-based ship detection methods: faster R-CNN [35] and YOLOv3 [36], as the benchmarks in our work. Wefirst utilize BNN to pretrain the feature extraction module for the selected SAR ship detectors, namely, the ResNet50 [8] backbone for faster RCNN [35] and the Darknet53 [36] backbone for YOLOv3 [36], based on QXS-SAROPT with 14,000 image pairs as the training set and the remaining 6,000 image pairs as the testing set. Then, the pretrained model with better SAR feature embedding capabilities by common representation learning from optical and SAR images is used forfinetuning the corresponding SAR ship detector. 4.1. Dataset. Four benchmark SAR ship detection datasets are tested: AIR-SARShip-1.0 [37], AIR-SARShip-2.0 [38], HRSID [39], and SSDD [40]. The AIR-SARShip-1.0 dataset consists of 31 highresolution large-scale 3000 × 3000 images from the Gaofen3 satellite. 21 images are randomly selected as training and validation data, and the remaining 10 images are used for testing. The AIR-SARShip-2.0 dataset includes 300 images of size 1000 × 1000 with resolutions ranging from 1 m to 5 m from the Gaofen-3 satellite. 210 images are randomly selected as training and validation data, and the remaining 90 images are used for testing. Images in AIR-SARShip-1.0 and AIR-SARShip-2.0 datasets are cropped into 512 × 512 pixels with a 0:5 overlap. The HRSID dataset contains 5604 SAR images of size 800 × 800 and is divided into training and testing set at a ratio of 13 : 7. The resolutions of images in the SSDD dataset range from 1 m to 15 m, and 1160 images are divided into 928 images for training and 232 images for testing. For faster R-CNN [35], we directly input each image of AIR-SARShip-1.0, AIR-SARShip-2.0, and HRSID into the 1. Select three SAR images acquired by the Gaofen-3 satellite [31], which contain rich land cover types such as inland, offshore, and mountains. The spatial resolution of SAR imagery is 1m × 1m for each pixel 2. Download the optical images of the corresponding area from Google Earth [32] with a spatial resolution of 1m × 1m 3. Cut the whole optical-SAR image pair into several subregion image pairs according to the complexity of land coverage. After that, we can register the subregion image pairs separately instead of directly registering the whole image pair 4. Locate matching points of the subregion optical-SAR image pairs manually, which are selected as the geometrically invariant corner points of buildings, ships, roads, etc 5. Use an existing automatic image registration software to register the subregion optical-SAR image pairs. Optical imagery is registered to thefixed SAR image through the bilinear interpolation method 6. Crop the registered subregion optical-SAR image pairs into small patches of 256 × 256 pixels with 20% overlap between adjacent patches 7. Double checked all image patches manually to ensure that every image contains meaningful information and texture. Remove indistinguishable orflawed images, such as images with similar scenes, texture-less sea, or visible mosaicking seamlines Algorithm 1: Process of constructing QXS-SAROPT dataset. Figure 5: Some exemplary patch-pairs from the QXS-SAROPT dataset. Top part: Google Earth RGB image patches, bottom part: Gaofen-3 SAR image patches. Table 1: Results of optical-SAR image matching using BNN [33] on QXS-SAROPT. Backbone Accuracy Precision Recall ResNet50 [8] 0.829 0.748 0.993 Darknet53 [36] 0.828 0.746 0.995 Space: Science & Technology 5
第17页
7 extraction module of BNN for the image modality to be interpreted is used as the pretained model for feature embedding of the specific crossmodal intelligent interpretation task. In this way, plentiful complementary features transferred from images of the other modality during the learning of common feature space can be used to enhance the feature embeddings to be interpreted. Benefited from the enhanced feature embeddings, our framework can effectively improve the interpretation performance of crossmodal SAR/optical images. Take SAR ship detection as an example. SAR ship detection in complex scenes is a great challenging task. And CNN-based SAR ship detection methods have drawn considerable attention because of the powerful feature embedding ability. Due to the scarce labeled SAR images, the pretraining technique is adopted to support these CNNbased SAR ship detectors. As SAR completely different from optical images, directly leveraging ImageNet [34] pretraining is hardly to obtain a good ship detector. However, our proposed framework can transfer rich texture features from optical images to SAR images to obtain a specific feature extraction model with better SAR feature embedding capabilities. Specifically, our proposed framework resorts to a SAR feature embedding operator from common representation learning based on the optical-SAR image matching task using BNN. 3. QXS-SAROPT Dataset To fully exploit the relevance of optical and SAR images and verify the effectiveness of our proposed framework, a large perfectly aligned optical-SAR image dataset with diverse scenes of a high resolution is in need. Considering the existing optical-SAR image matching dataset either lack scene diversity or has a low resolution, we have published the QXS-SAROPT dataset, which contains 20,000 pairs of SAR and optical image patches with a high resolution of 1 m extracted from multiple Gaofen-3 and Google Earth [32] scenes. As far as we know, QXS-SAROPT is thefirst dataset to provide high-resolution 1m × 1m coregistered SAR and optical satellite image patches covering over three big port cities in the world: San Diego, Shanghai, and Qingdao. The coverage of these images is shown in Figure 4. Algorithm 1 shows the procedure for the QXS-SAROPT dataset construction. Finally, 20,000 high-quality image patch pairs are preserved in our dataset, some of which are shown in Figure 5 for examples. 4. Experiments To verify the effectiveness and superiority of our BNN-based optical-SAR image joint intelligent interpretation framework, Object detection SAR image Optical image (a) (b) Common feature space SAR image Optical image SAR image Common feature space Figure 3: Two usage scenarios of the proposed BNN-based optical-SAR image joint intelligent interpretation framework. (a) Example illustration of the proposed framework applied to optical and SAR fusion intelligent interpretation. (b) Example illustration of the proposed framework applied to optical to SAR cross-modal intelligent interpretation. (a) (b) (c) Figure 4: Coverage of the SAR images of Gaofen-3 for constructing the QXS-SAROPT dataset. (a) SanDiego. (b) Shanghai. (c) Qingdao. 4 Space: Science & Technology the framework is applied to one typical optical-to-SAR crossmodal object detection task. We conduct optical-to-SAR crossmodal object detection tasks on four benchmark SAR ship detection datasets. We select two representative CNN-based ship detection methods: faster R-CNN [35] and YOLOv3 [36], as the benchmarks in our work. Wefirst utilize BNN to pretrain the feature extraction module for the selected SAR ship detectors, namely, the ResNet50 [8] backbone for faster RCNN [35] and the Darknet53 [36] backbone for YOLOv3 [36], based on QXS-SAROPT with 14,000 image pairs as the training set and the remaining 6,000 image pairs as the testing set. Then, the pretrained model with better SAR feature embedding capabilities by common representation learning from optical and SAR images is used forfinetuning the corresponding SAR ship detector. 4.1. Dataset. Four benchmark SAR ship detection datasets are tested: AIR-SARShip-1.0 [37], AIR-SARShip-2.0 [38], HRSID [39], and SSDD [40]. The AIR-SARShip-1.0 dataset consists of 31 highresolution large-scale 3000 × 3000 images from the Gaofen3 satellite. 21 images are randomly selected as training and validation data, and the remaining 10 images are used for testing. The AIR-SARShip-2.0 dataset includes 300 images of size 1000 × 1000 with resolutions ranging from 1 m to 5 m from the Gaofen-3 satellite. 210 images are randomly selected as training and validation data, and the remaining 90 images are used for testing. Images in AIR-SARShip-1.0 and AIR-SARShip-2.0 datasets are cropped into 512 × 512 pixels with a 0:5 overlap. The HRSID dataset contains 5604 SAR images of size 800 × 800 and is divided into training and testing set at a ratio of 13 : 7. The resolutions of images in the SSDD dataset range from 1 m to 15 m, and 1160 images are divided into 928 images for training and 232 images for testing. For faster R-CNN [35], we directly input each image of AIR-SARShip-1.0, AIR-SARShip-2.0, and HRSID into the 1. Select three SAR images acquired by the Gaofen-3 satellite [31], which contain rich land cover types such as inland, offshore, and mountains. The spatial resolution of SAR imagery is 1m × 1m for each pixel 2. Download the optical images of the corresponding area from Google Earth [32] with a spatial resolution of 1m × 1m 3. Cut the whole optical-SAR image pair into several subregion image pairs according to the complexity of land coverage. After that, we can register the subregion image pairs separately instead of directly registering the whole image pair 4. Locate matching points of the subregion optical-SAR image pairs manually, which are selected as the geometrically invariant corner points of buildings, ships, roads, etc 5. Use an existing automatic image registration software to register the subregion optical-SAR image pairs. Optical imagery is registered to thefixed SAR image through the bilinear interpolation method 6. Crop the registered subregion optical-SAR image pairs into small patches of 256 × 256 pixels with 20% overlap between adjacent patches 7. Double checked all image patches manually to ensure that every image contains meaningful information and texture. Remove indistinguishable orflawed images, such as images with similar scenes, texture-less sea, or visible mosaicking seamlines Algorithm 1: Process of constructing QXS-SAROPT dataset. Figure 5: Some exemplary patch-pairs from the QXS-SAROPT dataset. Top part: Google Earth RGB image patches, bottom part: Gaofen-3 SAR image patches. Table 1: Results of optical-SAR image matching using BNN [33] on QXS-SAROPT. Backbone Accuracy Precision Recall ResNet50 [8] 0.829 0.748 0.993 Darknet53 [36] 0.828 0.746 0.995 Space: Science & Technology 5
第18页
8 network and resize images in SSDD to 1000 × 600 pixels. As for YOLOv3 [36], all the images are resized to 416 × 416 pixels. Because of the necessity of the multiscaling training strategy for YOLOv3, no data augmentation except for scaling is applied. 4.2. Parameter Settings 4.2.1. Optical-SAR Image Matching. The BNN model with ResNet50 [8] and Darknet53 [36] as backbone is both trained in SGD for 200 epochs with a batch size of 20. The initial learning rate is set as 0.01 and then divided by a factor of 2 at the 30th and 100th epochs. The SAR and optical images are encoded into a 50-dimensional feature representation subspace. The ratio of positive and negative samples is set as 1 : 1 and the adjusting factorα = 1. 4.2.2. SAR Ship Detectors. For the faster R-CNN [35] benchmark, all models are trained with SGD for 14 epochs with 0.0001 weight decay and 0.9 momentum, and the batch size is set to 8. The initial learning rate is 0.02 and is then divided by 10 at the 8th and 12th epochs. For the YOLOv3 [36] benchmark, all models are trained with SGD for 240 epochs with 12 images per minibatch. The initial learning rate is set as 0.001 and is then divided by 10 at the 160th and 200th epochs. The IoU threshold is set as 0.5 when training and testing for rigorousfiltering of the bounding boxes with low precision. Warm-up [8] is used at thefirst 500 iterations during the training stage to avoid gradient explosion. The same settings are applied for all experiments for a fair comparison. 4.3. Results 4.3.1. Optical-SAR Image Matching. Table 1 shows the results of optical-SAR image matching using BNN [33] on the QXS-SAROPT dataset, which suggests that BNN achieves an outstanding performance on both ResNet50 [8] and Darknet53 [36] backbone. Specifically, the matching accuracy based on ResNet50 and Darknet53 reach up to 82:9% and 82:8%, respectively, demonstrating that BNN can learn useful common representations and well predict the relationship between SAR and optical images. Image pairs with different matching results by BNN are shown in Figure 6. To explore the relationship between the training set size and matching results, we randomly select 4000 and 8000 optical-SAR image pairs as the training sets to train BNN on ResNet50. Table 2 shows the results of three sizes of training sets, which indicates that BNN can learn a good common representation even with a small number of training image pairs, and more training data can lead to better matching results. Besides, we show the accuracy, precision, and recall curves of BNN with 8000 image pairs as the training set in Figure 7, which show the convergence process of BNN. 4.3.2. SAR Ship Detection. Table 3 shows the average precision (AP) of SAR ship detection results on four ship detection datasets using ImageNet pretraining-based SAR ship detector (ImageNet-SSD) and our BNN-based SAR ship detector (BNN-based-SSD) pretrained on QXS-SAROPT. As shown in Table 3, compared with ImageNet-SSD, the AP of detection results is generally improved by BNNbased-SSD. Especially on SAR ship detection dataset AIRSARShip-1.0 [37], 1.32% and 1.24% performance improvement can be achieved using two-stage detection benchmark: faster R-CNN [35] and one-stage detection benchmark: YOLOv3 [36], respectively. Average precision of ImageNetSSD and BNN-based-SSD during the training on the test set of HRSID and AIR-SARShip-2.0 with two detectors are displayed in Figure 8. Taking YOLOv3 in AIR-SARShip2.0 dataset as an example, BNN-based-SSD achieves higher average precision than ImageNet-SSD during the whole training process, indicating the significant improvement of our BNN-based-SSD. Similar phenomena are also presented on other datasets for both benchmarks, demonstrating the superiority of our BNN-based optical-SAR images joint intelligent interpretation framework. All these improved performances prove that our framework can well enhance (a) True positive (b) True negative (c) False positive (d) False negative Figure 6: Image pairs with different matching results by BNN [33] for a test subset of QXS-SAROPT. The image pairs correctly predicted as matched and unmatched pairs are considered as the true positive and true negative, respectively. False positive and false negative are, respectively, unmatched and matched images pairs that are wrongly predicted. Table 2: Results of optical-SAR image matching using BNN [33] on different sizes of training sets on ResNet50. Training set size Accuracy Precision Recall 4000 (25%) 0.731 0.656 0.972 8000 (50%) 0.793 0.714 0.977 16000 (100%) 0.829 0.748 0.993 6 Space: Science & Technology the feature extraction capability of SAR ship detectors by common representation learning utilizing BNN and thus boosting ship detection in SAR images even with no additional annotation information of ships. To qualitatively compare these two methods, we visualize some detection results in Figure 9, which shows that our BNN-based-SSD clearly outperforms ImageNet-SSD and significantly reduces the missed detections and false alarms. 5. Conclusion and Future Work In this paper, we propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, which can effectively improve the generic feature extraction capability of optical and SAR images by mining their feature correlation through matching tasks with BNN, and then improve the accuracy and application scenarios of specific optical-SAR image joint intelligent interpretation tasks. In order to fully exploit the correlation between optical and SAR images and ensure the great representation learning ability of BNN, we publish the QXSSAROPT dataset containing 20,000 optical-SAR patch pairs from multiple scenes of a high resolution of 1 meter. Experimental results on the optical-to-SAR crossmodal object detection task demonstrate the effectiveness and superiority of our framework. It is noted that based on the QXSSAROPT dataset, our framework can achieve up to 96% high accuracy in SAR ship detection. This research is in its early stage. In the future, we will consider exploring the performance of the proposed framework on optical-SAR fusion intelligent interpretation tasks, such as classification of land use and land cover and building segmentation. To support the research in intelligent interpretation fusing optical-SAR data, we will add label annotations and positions for scenes/objects of interest to every patch pair of the QXS-SAROPT dataset. In addition, to further explore the potential value of the QXS-SAROPT dataset, we are going to release an improved version of the dataset in the future, which will cover more land areas with versatile scenes and different sized patch pairs suitable for various optical-SAR data fusion tasks. At a more macroscopic level, there are plentiful aspects that deserve deeper investigation. Currently, our approach to interpreting multimodal remote sensing images is verified by experiments on the ground. However, the onboard processing of remote sensing images will be a trend in the future. Unfortunately, running deep learning models tends to be a high-power consumption process, let alone the tight constraints of onboard memory and computing resources. In 1.0 0.9 0.8 0.7 0.6 0.4 0.5 Accuracy 25 50 75 100 125 150 175 200 Epoch (a) (b) (c) Train Test 1.0 0.9 0.8 0.7 0.6 0.4 0.5 25 50 75 100 125 150 175 200 Epoch Precision 1.0 0.9 0.8 0.7 0.6 0.4 0.5 25 50 75 100 125 150 175 200 Epoch Recall Figure 7: Training process of optical-SAR image matching using BNN on 8000 image pairs in QXS-SAROPT dataset. (a) Accuracy. (b) Precision. (c) Recall. Table 3: Average precision of SAR ship detection results on four ship detection datasets using ImageNet-SSD pretrained on ImageNet and BNN-based-SSD pretrained on QXS-SAROPT. (a) AIR-SARShip-1.0 [37] HRSID [39] Faster R-CNN [35] YOLOv3 [36] Faster R-CNN [35] YOLOv3 [36] ImageNet-SSD 0.8720 0.8712 0.8878 0.8364 BNN-based-SSD 0.8852 0.8836 0.8932 0.8454 (b) AIR-SARShip-2.0 [38] SSDD [40] Faster R-CNN [35] YOLOv3 [36] Faster R-CNN [35] YOLOv3 [36] ImageNet-SSD 0.8487 0.8300 0.9624 0.9447 BNN-based-SSD 0.8582 0.8470 0.9679 0.9468 Space: Science & Technology 7
第19页
9 network and resize images in SSDD to 1000 × 600 pixels. As for YOLOv3 [36], all the images are resized to 416 × 416 pixels. Because of the necessity of the multiscaling training strategy for YOLOv3, no data augmentation except for scaling is applied. 4.2. Parameter Settings 4.2.1. Optical-SAR Image Matching. The BNN model with ResNet50 [8] and Darknet53 [36] as backbone is both trained in SGD for 200 epochs with a batch size of 20. The initial learning rate is set as 0.01 and then divided by a factor of 2 at the 30th and 100th epochs. The SAR and optical images are encoded into a 50-dimensional feature representation subspace. The ratio of positive and negative samples is set as 1 : 1 and the adjusting factorα = 1. 4.2.2. SAR Ship Detectors. For the faster R-CNN [35] benchmark, all models are trained with SGD for 14 epochs with 0.0001 weight decay and 0.9 momentum, and the batch size is set to 8. The initial learning rate is 0.02 and is then divided by 10 at the 8th and 12th epochs. For the YOLOv3 [36] benchmark, all models are trained with SGD for 240 epochs with 12 images per minibatch. The initial learning rate is set as 0.001 and is then divided by 10 at the 160th and 200th epochs. The IoU threshold is set as 0.5 when training and testing for rigorousfiltering of the bounding boxes with low precision. Warm-up [8] is used at thefirst 500 iterations during the training stage to avoid gradient explosion. The same settings are applied for all experiments for a fair comparison. 4.3. Results 4.3.1. Optical-SAR Image Matching. Table 1 shows the results of optical-SAR image matching using BNN [33] on the QXS-SAROPT dataset, which suggests that BNN achieves an outstanding performance on both ResNet50 [8] and Darknet53 [36] backbone. Specifically, the matching accuracy based on ResNet50 and Darknet53 reach up to 82:9% and 82:8%, respectively, demonstrating that BNN can learn useful common representations and well predict the relationship between SAR and optical images. Image pairs with different matching results by BNN are shown in Figure 6. To explore the relationship between the training set size and matching results, we randomly select 4000 and 8000 optical-SAR image pairs as the training sets to train BNN on ResNet50. Table 2 shows the results of three sizes of training sets, which indicates that BNN can learn a good common representation even with a small number of training image pairs, and more training data can lead to better matching results. Besides, we show the accuracy, precision, and recall curves of BNN with 8000 image pairs as the training set in Figure 7, which show the convergence process of BNN. 4.3.2. SAR Ship Detection. Table 3 shows the average precision (AP) of SAR ship detection results on four ship detection datasets using ImageNet pretraining-based SAR ship detector (ImageNet-SSD) and our BNN-based SAR ship detector (BNN-based-SSD) pretrained on QXS-SAROPT. As shown in Table 3, compared with ImageNet-SSD, the AP of detection results is generally improved by BNNbased-SSD. Especially on SAR ship detection dataset AIRSARShip-1.0 [37], 1.32% and 1.24% performance improvement can be achieved using two-stage detection benchmark: faster R-CNN [35] and one-stage detection benchmark: YOLOv3 [36], respectively. Average precision of ImageNetSSD and BNN-based-SSD during the training on the test set of HRSID and AIR-SARShip-2.0 with two detectors are displayed in Figure 8. Taking YOLOv3 in AIR-SARShip2.0 dataset as an example, BNN-based-SSD achieves higher average precision than ImageNet-SSD during the whole training process, indicating the significant improvement of our BNN-based-SSD. Similar phenomena are also presented on other datasets for both benchmarks, demonstrating the superiority of our BNN-based optical-SAR images joint intelligent interpretation framework. All these improved performances prove that our framework can well enhance (a) True positive (b) True negative (c) False positive (d) False negative Figure 6: Image pairs with different matching results by BNN [33] for a test subset of QXS-SAROPT. The image pairs correctly predicted as matched and unmatched pairs are considered as the true positive and true negative, respectively. False positive and false negative are, respectively, unmatched and matched images pairs that are wrongly predicted. Table 2: Results of optical-SAR image matching using BNN [33] on different sizes of training sets on ResNet50. Training set size Accuracy Precision Recall 4000 (25%) 0.731 0.656 0.972 8000 (50%) 0.793 0.714 0.977 16000 (100%) 0.829 0.748 0.993 6 Space: Science & Technology the feature extraction capability of SAR ship detectors by common representation learning utilizing BNN and thus boosting ship detection in SAR images even with no additional annotation information of ships. To qualitatively compare these two methods, we visualize some detection results in Figure 9, which shows that our BNN-based-SSD clearly outperforms ImageNet-SSD and significantly reduces the missed detections and false alarms. 5. Conclusion and Future Work In this paper, we propose a bridge neural network- (BNN-) based optical-SAR image joint intelligent interpretation framework, which can effectively improve the generic feature extraction capability of optical and SAR images by mining their feature correlation through matching tasks with BNN, and then improve the accuracy and application scenarios of specific optical-SAR image joint intelligent interpretation tasks. In order to fully exploit the correlation between optical and SAR images and ensure the great representation learning ability of BNN, we publish the QXSSAROPT dataset containing 20,000 optical-SAR patch pairs from multiple scenes of a high resolution of 1 meter. Experimental results on the optical-to-SAR crossmodal object detection task demonstrate the effectiveness and superiority of our framework. It is noted that based on the QXSSAROPT dataset, our framework can achieve up to 96% high accuracy in SAR ship detection. This research is in its early stage. In the future, we will consider exploring the performance of the proposed framework on optical-SAR fusion intelligent interpretation tasks, such as classification of land use and land cover and building segmentation. To support the research in intelligent interpretation fusing optical-SAR data, we will add label annotations and positions for scenes/objects of interest to every patch pair of the QXS-SAROPT dataset. In addition, to further explore the potential value of the QXS-SAROPT dataset, we are going to release an improved version of the dataset in the future, which will cover more land areas with versatile scenes and different sized patch pairs suitable for various optical-SAR data fusion tasks. At a more macroscopic level, there are plentiful aspects that deserve deeper investigation. Currently, our approach to interpreting multimodal remote sensing images is verified by experiments on the ground. However, the onboard processing of remote sensing images will be a trend in the future. Unfortunately, running deep learning models tends to be a high-power consumption process, let alone the tight constraints of onboard memory and computing resources. In 1.0 0.9 0.8 0.7 0.6 0.4 0.5 Accuracy 25 50 75 100 125 150 175 200 Epoch (a) (b) (c) Train Test 1.0 0.9 0.8 0.7 0.6 0.4 0.5 25 50 75 100 125 150 175 200 Epoch Precision 1.0 0.9 0.8 0.7 0.6 0.4 0.5 25 50 75 100 125 150 175 200 Epoch Recall Figure 7: Training process of optical-SAR image matching using BNN on 8000 image pairs in QXS-SAROPT dataset. (a) Accuracy. (b) Precision. (c) Recall. Table 3: Average precision of SAR ship detection results on four ship detection datasets using ImageNet-SSD pretrained on ImageNet and BNN-based-SSD pretrained on QXS-SAROPT. (a) AIR-SARShip-1.0 [37] HRSID [39] Faster R-CNN [35] YOLOv3 [36] Faster R-CNN [35] YOLOv3 [36] ImageNet-SSD 0.8720 0.8712 0.8878 0.8364 BNN-based-SSD 0.8852 0.8836 0.8932 0.8454 (b) AIR-SARShip-2.0 [38] SSDD [40] Faster R-CNN [35] YOLOv3 [36] Faster R-CNN [35] YOLOv3 [36] ImageNet-SSD 0.8487 0.8300 0.9624 0.9447 BNN-based-SSD 0.8582 0.8470 0.9679 0.9468 Space: Science & Technology 7
第20页
10 Ground truth ImageNet BNN Figure 9: Visualization of comparison results between ImageNet-SSD and BNN-based SSD. The red and green rectanglesfigure out the ground truth and correctly predicted boxes. The orange and yellow circles denote the false alarms and missing detection. Average precision Epoch (a) (b) (c) (d) 1.00 0.95 0.90 0.85 0.80 0.70 0.75 2 4 121086 14 HRSID for faster R-CNN 1.0 0.8 0.6 0.4 0.0 0.2 Average precision Epoch 2 4 121086 14 AIR-SARship-2.0 for faster R-CNN 1.0 0.9 0.8 0.7 0.5 0.6 0 50 100 150 200 ImageNet-SSD BNN-based-SSD Average precision Epoch HRSID for YOLOv3 1.0 0.8 0.6 0.4 0.0 0.2 0 50 100 150 200 Average precision Epoch AIR-SARship-2.0 for YOLOv3 Figure 8: Average precision of ImageNet-SSD and BNN-based-SSD during the training on test set of HRSID and AIR-SARShip-2.0 with faster R-CNN and YOLOv3. 8 Space: Science & Technology this case, deep learning model compression is an effective and necessary technique to achieve onboard processing in our future work. The purpose of model compression is to achieve a model with fewer parameters, calculation amount, and less RAM to run without significantly diminished accuracy. Popular model compression methods include pruning [41], quantization [42], low-rank approximation and sparsity [43], and knowledge distillation [44, 45]. Furthermore, the formation of SAR images from echoes is thefirst inevitable step of SAR data processing nowadays, based on the algorithms such as back projection [46], compressed sensing [47], or signal processing. Therefore, the SAR application pipeline consists of multiple operations and varieties of complex calculations. In our future work, we will attempt to develop a deep learning framework that performs an integrating SAR processing workflow end to end, from the reflected echoes to the interpretation results. This will help to reduce the complexity of the onboard processor and further improve the processing efficiency. Data Availability The QXS-SAROPT dataset released by this work is publicly available at https://github.com/yaoxu008/QXS-SAROPT under open access license CCBY. Four SAR ship detection datasets used in this paper are publicly available. AIR-SARShip-1.0 and AIR-SARShip-2.0 can be accessed at http://radars.ie.ac.cn/ web/data/getData?dataType=SARDataset. HRSID is available at https://github.com/chaozhong2010/HRSID. SSDD can be downloaded at https://github.com/CAESAR-Radi/SAR-ShipDataset. Conflicts of Interest All authors declare no possible conflicts of interests. Authors’ Contributions X. Xiang and M. Huang conceived the idea of this study and supervised the study. M. Huang, X. Yao, L. Qian, and W. Bao conducted the experiments. M. Huang and L. Qian performed data analysis. M. Huang, X. Yao, and L. Qian contributed to the writing of the manuscript. L. Qian, W. Shi, Y. Zhang, N. Wang, and X. Liu participated in the construction of the QXS-SAROPT dataset. Meiyu Huang, Yao Xu, and Lixin Qian contributed equally to this work. Acknowledgments This is supported by the Beijing Nova Program of Science and Technology under Grant Z191100001119129 and the National Natural Science Foundation of China 61702520. References [1] L. Zhang, L. Zhang, and B. Du,“Deep learning for remote sensing data: a technical tutorial on the state of the art,” IEEE Geoscience and Remote Sensing Magazine, vol. 4, no. 2, pp. 22–40, 2016. [2] X. X. Zhu, D. Tuia, L. Mou et al.,“Deep learning in remote sensing: a comprehensive review and list of resources,” IEEE Geoscience and Remote Sensing Magazine, vol. 5, no. 4, pp. 8–36, 2017. [3] J. E. Ball, D. T. Anderson, and C. S. Chan,“Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community,” Journal of Applied Remote Sensing, vol. 11, no. 4, 2017. [4] G. Tsagkatakis, A. Aidini, K. Fotiadou, M. Giannopoulos, A. Pentari, and P. Tsakalides,“Survey of deep-learning approaches for remote sensing observation enhancement,” Sensors, vol. 19, no. 18, article 3929, 2019. [5] T. N. Sainath, A.-r. Mohamed, B. Kingsbury, and B. Ramabhadran,“Deep Convolutional Neural Networks for Lvcsr,” in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 8614–8618, Vancouver, BC, Canada, 2013. [6] K. Simonyan and A. Zisserman,“Very deep convolutional networks for large-scale image recognition,”in International Conference on Learning Representations, San Diego, CA, USA, 2015. [7] J. Schmidhuber,“Deep learning in neural networks: An overview,” Neural Networks, vol. 61, pp. 85–117, 2015. [8] K. He, X. Zhang, S. Ren, and J. Sun,“Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, Las Vegas, NV, USA, 2016. [9] M. Schmitt and X. X. Zhu,“Data fusion and remote sensing: an ever-growing relationship,” IEEE Geoscience and Remote Sensing Magazine, vol. 4, no. 4, pp. 6–23, 2016. [10] Z. Zhang, G. Vosselman, M. Gerke, D. Tuia, and M. Y. Yang, “Change detection between multimodal remote sensing data using siamese cnn,” 2018, https://arxiv.org/abs/1807.09562. [11] P. Feng, Y. Lin, J. Guan et al.,“Embranchment cnn based local climate zone classification using sar and multispectral remote sensing data,” in IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, pp. 6344–6347, Yokohama, Japan, 2019. [12] Z. Zhang, G. Vosselman, M. Gerke, C. Persello, D. Tuia, and M. Y. Yang,“Detecting building changes between airborne laser scanning and photogrammetric data,” Remote sensing, vol. 11, no. 20, article 2417, 2019. [13] M. Schmitt, F. Tupin, and X. X. Zhu,“Fusion of sar and optical remote sensing data–challenges and recent trends,” in 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 5458–5461, Fort Worth, TX, USA, 2017. [14] M. Schmitt, L. Hughes, and X. Zhu,“The sen1-2 dataset for deep learning in sar-optical data FUSION,” Remote Sensing & Spatial Information Sciences, vol. IV-1, no. 1, pp. 141–146, 2018. [15] Q. Feng, J. Yang, D. Zhu et al.,“Integrating multitemporal sentinel-1/2 data for coastal land cover classification using a multibranch convolutional neural network: a case of the yellow river delta,” Remote Sensing, vol. 11, no. 9, article 1006, 2019. [16] S. C. Kulkarni and P. P. Rege,“Pixel level fusion techniques for sar and optical images: a review,” Information Fusion, vol. 59, pp. 13–29, 2020. [17] X. Li, L. Lei, Y. Sun, M. Li, and G. Kuang,“Multimodal bilinear fusion network with Second-Order attention-based channel selection for land cover classification,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 1011–1026, 2020. Space: Science & Technology 9
第21页
11 Ground truth ImageNet BNN Figure 9: Visualization of comparison results between ImageNet-SSD and BNN-based SSD. The red and green rectanglesfigure out the ground truth and correctly predicted boxes. The orange and yellow circles denote the false alarms and missing detection. Average precision Epoch (a) (b) (c) (d) 1.00 0.95 0.90 0.85 0.80 0.70 0.75 2 4 121086 14 HRSID for faster R-CNN 1.0 0.8 0.6 0.4 0.0 0.2 Average precision Epoch 2 4 121086 14 AIR-SARship-2.0 for faster R-CNN 1.0 0.9 0.8 0.7 0.5 0.6 0 50 100 150 200 ImageNet-SSD BNN-based-SSD Average precision Epoch HRSID for YOLOv3 1.0 0.8 0.6 0.4 0.0 0.2 0 50 100 150 200 Average precision Epoch AIR-SARship-2.0 for YOLOv3 Figure 8: Average precision of ImageNet-SSD and BNN-based-SSD during the training on test set of HRSID and AIR-SARShip-2.0 with faster R-CNN and YOLOv3. 8 Space: Science & Technology this case, deep learning model compression is an effective and necessary technique to achieve onboard processing in our future work. The purpose of model compression is to achieve a model with fewer parameters, calculation amount, and less RAM to run without significantly diminished accuracy. Popular model compression methods include pruning [41], quantization [42], low-rank approximation and sparsity [43], and knowledge distillation [44, 45]. Furthermore, the formation of SAR images from echoes is thefirst inevitable step of SAR data processing nowadays, based on the algorithms such as back projection [46], compressed sensing [47], or signal processing. Therefore, the SAR application pipeline consists of multiple operations and varieties of complex calculations. In our future work, we will attempt to develop a deep learning framework that performs an integrating SAR processing workflow end to end, from the reflected echoes to the interpretation results. This will help to reduce the complexity of the onboard processor and further improve the processing efficiency. Data Availability The QXS-SAROPT dataset released by this work is publicly available at https://github.com/yaoxu008/QXS-SAROPT under open access license CCBY. Four SAR ship detection datasets used in this paper are publicly available. AIR-SARShip-1.0 and AIR-SARShip-2.0 can be accessed at http://radars.ie.ac.cn/ web/data/getData?dataType=SARDataset. HRSID is available at https://github.com/chaozhong2010/HRSID. SSDD can be downloaded at https://github.com/CAESAR-Radi/SAR-ShipDataset. Conflicts of Interest All authors declare no possible conflicts of interests. Authors’ Contributions X. Xiang and M. Huang conceived the idea of this study and supervised the study. M. Huang, X. Yao, L. Qian, and W. Bao conducted the experiments. M. Huang and L. Qian performed data analysis. M. Huang, X. Yao, and L. Qian contributed to the writing of the manuscript. L. Qian, W. Shi, Y. Zhang, N. Wang, and X. Liu participated in the construction of the QXS-SAROPT dataset. Meiyu Huang, Yao Xu, and Lixin Qian contributed equally to this work. Acknowledgments This is supported by the Beijing Nova Program of Science and Technology under Grant Z191100001119129 and the National Natural Science Foundation of China 61702520. References [1] L. Zhang, L. Zhang, and B. Du,“Deep learning for remote sensing data: a technical tutorial on the state of the art,” IEEE Geoscience and Remote Sensing Magazine, vol. 4, no. 2, pp. 22–40, 2016. [2] X. X. Zhu, D. Tuia, L. Mou et al.,“Deep learning in remote sensing: a comprehensive review and list of resources,” IEEE Geoscience and Remote Sensing Magazine, vol. 5, no. 4, pp. 8–36, 2017. [3] J. E. Ball, D. T. Anderson, and C. S. Chan,“Comprehensive survey of deep learning in remote sensing: theories, tools, and challenges for the community,” Journal of Applied Remote Sensing, vol. 11, no. 4, 2017. [4] G. Tsagkatakis, A. Aidini, K. Fotiadou, M. Giannopoulos, A. Pentari, and P. Tsakalides,“Survey of deep-learning approaches for remote sensing observation enhancement,” Sensors, vol. 19, no. 18, article 3929, 2019. [5] T. N. Sainath, A.-r. Mohamed, B. Kingsbury, and B. Ramabhadran,“Deep Convolutional Neural Networks for Lvcsr,” in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 8614–8618, Vancouver, BC, Canada, 2013. [6] K. Simonyan and A. Zisserman,“Very deep convolutional networks for large-scale image recognition,”in International Conference on Learning Representations, San Diego, CA, USA, 2015. [7] J. Schmidhuber,“Deep learning in neural networks: An overview,” Neural Networks, vol. 61, pp. 85–117, 2015. [8] K. He, X. Zhang, S. Ren, and J. Sun,“Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, Las Vegas, NV, USA, 2016. [9] M. Schmitt and X. X. Zhu,“Data fusion and remote sensing: an ever-growing relationship,” IEEE Geoscience and Remote Sensing Magazine, vol. 4, no. 4, pp. 6–23, 2016. [10] Z. Zhang, G. Vosselman, M. Gerke, D. Tuia, and M. Y. Yang, “Change detection between multimodal remote sensing data using siamese cnn,” 2018, https://arxiv.org/abs/1807.09562. [11] P. Feng, Y. Lin, J. Guan et al.,“Embranchment cnn based local climate zone classification using sar and multispectral remote sensing data,” in IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, pp. 6344–6347, Yokohama, Japan, 2019. [12] Z. Zhang, G. Vosselman, M. Gerke, C. Persello, D. Tuia, and M. Y. Yang,“Detecting building changes between airborne laser scanning and photogrammetric data,” Remote sensing, vol. 11, no. 20, article 2417, 2019. [13] M. Schmitt, F. Tupin, and X. X. Zhu,“Fusion of sar and optical remote sensing data–challenges and recent trends,” in 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 5458–5461, Fort Worth, TX, USA, 2017. [14] M. Schmitt, L. Hughes, and X. Zhu,“The sen1-2 dataset for deep learning in sar-optical data FUSION,” Remote Sensing & Spatial Information Sciences, vol. IV-1, no. 1, pp. 141–146, 2018. [15] Q. Feng, J. Yang, D. Zhu et al.,“Integrating multitemporal sentinel-1/2 data for coastal land cover classification using a multibranch convolutional neural network: a case of the yellow river delta,” Remote Sensing, vol. 11, no. 9, article 1006, 2019. [16] S. C. Kulkarni and P. P. Rege,“Pixel level fusion techniques for sar and optical images: a review,” Information Fusion, vol. 59, pp. 13–29, 2020. [17] X. Li, L. Lei, Y. Sun, M. Li, and G. Kuang,“Multimodal bilinear fusion network with Second-Order attention-based channel selection for land cover classification,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 1011–1026, 2020. Space: Science & Technology 9
第22页
12 [18] W. Burger and M. J. Burge, Principles of Digital Image Processing: Core Algorithms, 2010, Springer Science & Business Media. [19] J. Walters-Williams and Y. Li,“Estimation of mutual information: a survey,” in Rough Sets and Knowledge Technology. RSKT 2009, Lecture Notes in Computer Science, P. Wen, Y. Li, L. Polkowski, Y. Yao, S. Tsumoto, and G. Wang, Eds., pp. 389–396, Springer, Berlin, Heidelberg, 2009. [20] S. Suri and P. Reinartz,“Mutual-information-based registration of terrasar-x and ikonos imagery in urban areas,” IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 2, pp. 939–949, 2010. [21] D. G. Lowe,“Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision, vol. 60, no. 2, pp. 91–110, 2004. [22] F. Dellinger, J. Delon, Y. Gousseau, J. Michel, and F. Tupin, “Sar-sift: a sift-like algorithm for sar images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 1, pp. 453–466, 2015. [23] Y. Ye and L. Shen,“Hopc: a novel similarity metric based on geometric structural properties for multi-modal remote sensing image MATCHING,” Remote Sensing and Spatial Information Sciences, vol. III-1, pp. 9–16, 2016. [24] Y. Ye, J. Shan, L. Bruzzone, and L. Shen,“Robust registration of multimodal remote sensing images based on structural similarity,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 5, pp. 2941–2958, 2017. [25] S. Zagoruyko and N. Komodakis,“Learning to compare image patches via convolutional neural networks,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4353–4361, Boston, MA, USA, 2015. [26] N. Merkle, W. Luo, S. Auer, R. Müller, and R. Urtasun, “Exploiting deep matching and sar data for the geolocalization accuracy improvement of optical satellite images,” Remote Sensing, vol. 9, no. 6, article 586, 2017. [27] L. Mou, M. Schmitt, Y. Wang, and X. X. Zhu,“A cnn for the identification of corresponding patches in sar and optical imagery of urban scenes,” in 2017 Joint Urban Remote Sensing Event (JURSE), pp. 1–4, Dubai, United Arab Emirates, 2017. [28] L. H. Hughes, M. Schmitt, L. Mou, Y. Wang, and X. X. Zhu, “Identifying corresponding patches in sar and optical images with a pseudo-siamese cnn,” IEEE Geoscience and Remote Sensing Letters, vol. 15, no. 5, pp. 784–788, 2018. [29] Y. Wang and X. X. Zhu,“The sarptical dataset for joint analysis of sar and optical image in dense urban area,” in IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 6840–6843, Valencia, Spain, 2018. [30] J. Shermeyer, D. Hogan, J. Brown et al.,“Spacenet 6: Multisensor all weather mapping dataset,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 196-197, Seattle, WA, USA, 2020. [31] Q. Zhang,“System design and key technologies of the gf-3 satellite,” Acta Geodaetica et Cartographica Sinica, vol. 46, no. 3, pp. 269–277, 2017. [32] https://earth.google.com/. [33] Y. Xu, X. Xiang, and M. Huang,“Task-driven common representation learning via bridge neural network,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, pp. 5573–5580, 2019. [34] O. Russakovsky, J. Deng, H. Su et al.,“Imagenet large scale visual recognition challenge,” International Journal of Computer Vision, vol. 115, no. 3, pp. 211–252, 2015. [35] S. Ren, K. He, R. Girshick, and J. Sun,“Faster r-cnn: towards real-time object detection with region proposal networks,” Advances in Neural Information Processing Systems, pp. 91–99, 2015. [36] J. Redmon and A. Farhadi,“Yolov3: an incremental improvement,” 2018, https://arxiv.org/abs/1804.02767. [37] S. Xian, W. Zhirui, S. Yuanrui, D. Wenhui, Z. Yue, and F. Kun, “Air-sarship–1.0: High resolution sar ship detection dataset,” Journal of Radars, vol. 8, no. 6, pp. 852–862, 2019. [38]“2020 gaofen challenge on automated high-resolution earth observation image interpretation,” 2020, http://en.sw.chreos .org. [39] S. Wei, X. Zeng, Q. Qu, M. Wang, H. Su, and J. Shi,“Hrsid: A high-resolution sar images dataset for ship detection and instance segmentation,” IEEE Access, vol. 8, pp. 120234– 120254, 2020. [40] J. Li, C. Qu, and J. Shao,“Ship Detection in Sar Images Based on an Improved faster r-Cnn,” in 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), pp. 1–6, Beijing, China, 2017. [41] S. Han, J. Pool, J. Tran, and W. J. Dally,“Learning both weights and connections for efficient neural network,” Advances in Neural Information Processing Systems, MIT Press, 2015. [42] S. Han, H. Mao, and W. J. Dally,“Deep compression: compressing deep neural networks with pruning, trained quantization and human coding,” in Proceedings of International Conference on Learning Representations, San Juan, Puerto Rico, 2016. [43] M. Jaderberg, A. Vedaldi, and A. Zisserman,“Speeding up Convolutional Neural Networks with Low Rank Expansions,” in Proceedings of the British Machine Vision Conference, University of Nottingham, UK, 2014. [44] C. Bucilua, R. Caruana, and A. Niculescumizil,“Model compression,” in Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, pp. 535–541, 2006. [45] G. Hinton, O. Vinyals, and J. Dean,“Distilling the knowledge in a neural network,” Advances in Neural Information Processing Systems, MIT Press, 2014. [46] L. A. Gorham and L. J. Moore,“Sar image formation toolbox for matlab, in Algorithms for Synthetic Aperture Radar Imagery XVII,” International Society for Optics and Photonics, vol. 7699, pp. 769–906, 2010. [47] R. Baraniuk and P. Steeghs,“Compressive radar imaging,” in 2007 IEEE Radar Conference, pp. 128–133, Waltham, MA, USA, 2007. 10 Space: Science & Technology Research Article Developing Prototype Simulants for Surface Materials and Morphology of Near Earth Asteroid 2016 HO3 Xiaojing Zhang , 1 Yuechen Luo,1,2 Yuan Xiao,1 Deyun Liu,3 Fan Guo,4 and Qian Guo5 1 Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100029, China 2 School of Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China 3 Beijing Spacecraft Manufacturing Factory Co., Ltd., Beijing 100094, China 4 Beijing Institute of Space System Engineering, Beijing 100094, China 5 Institute of Geology and Geophysics, Chinese Academy of Sciences, Beijing 100029, China Correspondence should be addressed to Xiaojing Zhang; zhangxiaojing@qxslab.cn Received 25 August 2021; Accepted 20 October 2021; Published 8 November 2021 Copyright © 2021 Xiaojing Zhang et al. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). There are a variety of applications for asteroid simulants in asteroid studies for science advances as well as technology maturation. For specific purpose, it usually requires purpose-specialized simulant. In this study, we designed and developed a set of prototype simulants as S-type asteroid surface materials analogue based on H, L, and LL ordinary chondrites’ mineralogy and terrestrial observations of near-earth asteroid 2016 HO3, which is the Chinese sample return mission target. These simulants are able to simulate morphology and reflectance characteristics of asteroid (469219) 2016 HO3 and, thus, to be used for engineering evaluation of the optical navigation system and the sampling device of the spacecraft during the mission phase. Meanwhile, these prototype simulants are easily to modify to reflect newfindings on the asteroid surface when the spacecraft makes proximate observations. 1. Introduction Asteroid simulants are needed in supporting space missions, developing near-earth asteroids in situ resource utilization technology, and developing planetary defense techniques. It is unrealistic to create simulants that replicate all characteristics of the target asteroids due to different geological processes on asteroids and Earth [1]. Therefore, purposespecialized simulant is more useful and practical for actual space missions that always have specific purposes, such as navigation experiment and sampling test [2]. The recent asteroid missions to Itokawa, Ryugu, and Bennu have acquired numerous high-resolution images of these asteroids. Observations show that small asteroids (less than kilometer-sized) are rubble piles, and up to tens of meter-sized boulders and submicron- to millimeter-sized regolith particles are exposed on their surface [3–5]. Such a characteristic brings unexpected challenges to spacecraft when selecting landing and sampling sites, as what Hayabusa 2 and OSIRIS-Rex have encountered. China will launch itsfirst asteroid mission in the coming few years aiming to first return samples back from a near-earth asteroid (469219) 2016 HO3 and then rendezvous with a main belt comet 133P/Elst-Pizarro or 311P/PANSTARRS [6]. Due to the very much unknown surface morphology and properties of 2016 HO3, it is highly likely that this mission will be facing great challenges during sampling operation. To reduce risks, evaluations should have been performed before the actual landing and sampling operations, which would ultimately provide valuable information about the surface condition. In this case, it is necessary to develop asteroid simulants that can simulate the morphology and reflectance characteristics of 2016 HO3. Meanwhile, such simulants for 2016 HO3 should be easily to modify to reflect new observations made by the spacecraft. There are previous attempts to developing simulated materials for asteroids, most of which are concentrating on C-type asteroids. These simulants are usually based on carbonaceous chondrites plus remote observations of targeted asteroids, including a series of prototype simulants based on carbonaceous chondrites [7], simulant HCCL-1 for Bennu [8], and a simplified Ryugu simulant [2]. Simulants AAAS Space: Science & Technology Volume 2021, Article ID 9874929, 6 pages https://doi.org/10.34133/2021/9874929
第23页
13 [18] W. Burger and M. J. Burge, Principles of Digital Image Processing: Core Algorithms, 2010, Springer Science & Business Media. [19] J. Walters-Williams and Y. Li,“Estimation of mutual information: a survey,” in Rough Sets and Knowledge Technology. RSKT 2009, Lecture Notes in Computer Science, P. Wen, Y. Li, L. Polkowski, Y. Yao, S. Tsumoto, and G. Wang, Eds., pp. 389–396, Springer, Berlin, Heidelberg, 2009. [20] S. Suri and P. Reinartz,“Mutual-information-based registration of terrasar-x and ikonos imagery in urban areas,” IEEE Transactions on Geoscience and Remote Sensing, vol. 48, no. 2, pp. 939–949, 2010. [21] D. G. Lowe,“Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision, vol. 60, no. 2, pp. 91–110, 2004. [22] F. Dellinger, J. Delon, Y. Gousseau, J. Michel, and F. Tupin, “Sar-sift: a sift-like algorithm for sar images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 1, pp. 453–466, 2015. [23] Y. Ye and L. Shen,“Hopc: a novel similarity metric based on geometric structural properties for multi-modal remote sensing image MATCHING,” Remote Sensing and Spatial Information Sciences, vol. III-1, pp. 9–16, 2016. [24] Y. Ye, J. Shan, L. Bruzzone, and L. Shen,“Robust registration of multimodal remote sensing images based on structural similarity,” IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 5, pp. 2941–2958, 2017. [25] S. Zagoruyko and N. Komodakis,“Learning to compare image patches via convolutional neural networks,” in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4353–4361, Boston, MA, USA, 2015. [26] N. Merkle, W. Luo, S. Auer, R. Müller, and R. Urtasun, “Exploiting deep matching and sar data for the geolocalization accuracy improvement of optical satellite images,” Remote Sensing, vol. 9, no. 6, article 586, 2017. [27] L. Mou, M. Schmitt, Y. Wang, and X. X. Zhu,“A cnn for the identification of corresponding patches in sar and optical imagery of urban scenes,” in 2017 Joint Urban Remote Sensing Event (JURSE), pp. 1–4, Dubai, United Arab Emirates, 2017. [28] L. H. Hughes, M. Schmitt, L. Mou, Y. Wang, and X. X. Zhu, “Identifying corresponding patches in sar and optical images with a pseudo-siamese cnn,” IEEE Geoscience and Remote Sensing Letters, vol. 15, no. 5, pp. 784–788, 2018. [29] Y. Wang and X. X. Zhu,“The sarptical dataset for joint analysis of sar and optical image in dense urban area,” in IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 6840–6843, Valencia, Spain, 2018. [30] J. Shermeyer, D. Hogan, J. Brown et al.,“Spacenet 6: Multisensor all weather mapping dataset,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 196-197, Seattle, WA, USA, 2020. [31] Q. Zhang,“System design and key technologies of the gf-3 satellite,” Acta Geodaetica et Cartographica Sinica, vol. 46, no. 3, pp. 269–277, 2017. [32] https://earth.google.com/. [33] Y. Xu, X. Xiang, and M. Huang,“Task-driven common representation learning via bridge neural network,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, pp. 5573–5580, 2019. [34] O. Russakovsky, J. Deng, H. Su et al.,“Imagenet large scale visual recognition challenge,” International Journal of Computer Vision, vol. 115, no. 3, pp. 211–252, 2015. [35] S. Ren, K. He, R. Girshick, and J. Sun,“Faster r-cnn: towards real-time object detection with region proposal networks,” Advances in Neural Information Processing Systems, pp. 91–99, 2015. [36] J. Redmon and A. Farhadi,“Yolov3: an incremental improvement,” 2018, https://arxiv.org/abs/1804.02767. [37] S. Xian, W. Zhirui, S. Yuanrui, D. Wenhui, Z. Yue, and F. Kun, “Air-sarship–1.0: High resolution sar ship detection dataset,” Journal of Radars, vol. 8, no. 6, pp. 852–862, 2019. [38]“2020 gaofen challenge on automated high-resolution earth observation image interpretation,” 2020, http://en.sw.chreos .org. [39] S. Wei, X. Zeng, Q. Qu, M. Wang, H. Su, and J. Shi,“Hrsid: A high-resolution sar images dataset for ship detection and instance segmentation,” IEEE Access, vol. 8, pp. 120234– 120254, 2020. [40] J. Li, C. Qu, and J. Shao,“Ship Detection in Sar Images Based on an Improved faster r-Cnn,” in 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), pp. 1–6, Beijing, China, 2017. [41] S. Han, J. Pool, J. Tran, and W. J. Dally,“Learning both weights and connections for efficient neural network,” Advances in Neural Information Processing Systems, MIT Press, 2015. [42] S. Han, H. Mao, and W. J. Dally,“Deep compression: compressing deep neural networks with pruning, trained quantization and human coding,” in Proceedings of International Conference on Learning Representations, San Juan, Puerto Rico, 2016. [43] M. Jaderberg, A. Vedaldi, and A. Zisserman,“Speeding up Convolutional Neural Networks with Low Rank Expansions,” in Proceedings of the British Machine Vision Conference, University of Nottingham, UK, 2014. [44] C. Bucilua, R. Caruana, and A. Niculescumizil,“Model compression,” in Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, pp. 535–541, 2006. [45] G. Hinton, O. Vinyals, and J. Dean,“Distilling the knowledge in a neural network,” Advances in Neural Information Processing Systems, MIT Press, 2014. [46] L. A. Gorham and L. J. Moore,“Sar image formation toolbox for matlab, in Algorithms for Synthetic Aperture Radar Imagery XVII,” International Society for Optics and Photonics, vol. 7699, pp. 769–906, 2010. [47] R. Baraniuk and P. Steeghs,“Compressive radar imaging,” in 2007 IEEE Radar Conference, pp. 128–133, Waltham, MA, USA, 2007. 10 Space: Science & Technology Research Article Developing Prototype Simulants for Surface Materials and Morphology of Near Earth Asteroid 2016 HO3 Xiaojing Zhang , 1 Yuechen Luo,1,2 Yuan Xiao,1 Deyun Liu,3 Fan Guo,4 and Qian Guo5 1 Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100029, China 2 School of Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China 3 Beijing Spacecraft Manufacturing Factory Co., Ltd., Beijing 100094, China 4 Beijing Institute of Space System Engineering, Beijing 100094, China 5 Institute of Geology and Geophysics, Chinese Academy of Sciences, Beijing 100029, China Correspondence should be addressed to Xiaojing Zhang; zhangxiaojing@qxslab.cn Received 25 August 2021; Accepted 20 October 2021; Published 8 November 2021 Copyright © 2021 Xiaojing Zhang et al. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). There are a variety of applications for asteroid simulants in asteroid studies for science advances as well as technology maturation. For specific purpose, it usually requires purpose-specialized simulant. In this study, we designed and developed a set of prototype simulants as S-type asteroid surface materials analogue based on H, L, and LL ordinary chondrites’ mineralogy and terrestrial observations of near-earth asteroid 2016 HO3, which is the Chinese sample return mission target. These simulants are able to simulate morphology and reflectance characteristics of asteroid (469219) 2016 HO3 and, thus, to be used for engineering evaluation of the optical navigation system and the sampling device of the spacecraft during the mission phase. Meanwhile, these prototype simulants are easily to modify to reflect newfindings on the asteroid surface when the spacecraft makes proximate observations. 1. Introduction Asteroid simulants are needed in supporting space missions, developing near-earth asteroids in situ resource utilization technology, and developing planetary defense techniques. It is unrealistic to create simulants that replicate all characteristics of the target asteroids due to different geological processes on asteroids and Earth [1]. Therefore, purposespecialized simulant is more useful and practical for actual space missions that always have specific purposes, such as navigation experiment and sampling test [2]. The recent asteroid missions to Itokawa, Ryugu, and Bennu have acquired numerous high-resolution images of these asteroids. Observations show that small asteroids (less than kilometer-sized) are rubble piles, and up to tens of meter-sized boulders and submicron- to millimeter-sized regolith particles are exposed on their surface [3–5]. Such a characteristic brings unexpected challenges to spacecraft when selecting landing and sampling sites, as what Hayabusa 2 and OSIRIS-Rex have encountered. China will launch itsfirst asteroid mission in the coming few years aiming to first return samples back from a near-earth asteroid (469219) 2016 HO3 and then rendezvous with a main belt comet 133P/Elst-Pizarro or 311P/PANSTARRS [6]. Due to the very much unknown surface morphology and properties of 2016 HO3, it is highly likely that this mission will be facing great challenges during sampling operation. To reduce risks, evaluations should have been performed before the actual landing and sampling operations, which would ultimately provide valuable information about the surface condition. In this case, it is necessary to develop asteroid simulants that can simulate the morphology and reflectance characteristics of 2016 HO3. Meanwhile, such simulants for 2016 HO3 should be easily to modify to reflect new observations made by the spacecraft. There are previous attempts to developing simulated materials for asteroids, most of which are concentrating on C-type asteroids. These simulants are usually based on carbonaceous chondrites plus remote observations of targeted asteroids, including a series of prototype simulants based on carbonaceous chondrites [7], simulant HCCL-1 for Bennu [8], and a simplified Ryugu simulant [2]. Simulants AAAS Space: Science & Technology Volume 2021, Article ID 9874929, 6 pages https://doi.org/10.34133/2021/9874929 Developing Prototype Simulants for Surface Materials and Morphology of Near Earth Asteroid 2016 HO3 Xiaojing Zhang,1 Yuechen Luo,1,2 Yuan Xiao,1 Deyun Liu,3 Fan Guo,4 and Qian Guo5 1 Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100029, China 2 School of Earth Sciences and Resources, China University of Geosciences, Beijing 100083, China 3 Beijing Spacecraft Manufacturing Factory Co., Ltd., Beijing 100094, China 4 Beijing Institute of Space System Engineering, Beijing 100094, China 5 Institute of Geology and Geophysics, Chinese Academy of Sciences, Beijing 100029, China Correspondence should be addressed to Xiaojing Zhang; zhangxiaojing@qxslab.cn Abstract: There are a variety of applications for asteroid simulants in asteroid studies for science advances as well as technology maturation. For specific purpose, it usually requires purpose-specialized simulant. In this study, we designed and developed a set of prototype simulants as S-type asteroid surface materials analogue based on H, L, and LL ordinary chondrites’ mineralogy and terrestrial observations of near-earth asteroid 2016 HO3, which is the Chinese sample return mission target. These simulants are able to simulate morphology and reflectance characteristics of asteroid (469219) 2016 HO3 and, thus, to be used for engineering evaluation of the optical navigation system and the sampling device of the spacecraft during the mission phase. Meanwhile, these prototype simulants are easily to modify to reflect new findings on the asteroid surface when the spacecraft makes proximate observations.
第24页
14 for S-type asteroid are relatively less reported. A new Itokawa asteroid regolith simulant (called IRS-1) as an S-type asteroid was developed for China’s upcoming asteroid exploration [9]. However, this simulant only containsfine grains ranging from submillimeter to centimeter, without larger-sized boulders. This study designed and developed the morphology equivalent prototype simulants for 2016 HO3 that are with particle sizes ranging from micrometer to at least tens of centimeters. The goal of these simulants is to simulate a possible rubble pile morphology and the reflectance feature of the surface of 2016 HO3 so that they can serve practical uses during the mission preparation and operation phases, such as test and evaluation of navigation system and sampling device before the actual touchdown or landing operation. Furthermore, when the spacecraft’s initial observation reveals a different size distribution of 2016 HO3 surface than our estimation, which is always the case, our simulants can be easily modified to reflect the newfindings. 2. Observation of (469219) 2016 HO3 The near-earth asteroid (469219) 2016 HO3 was found on 27th of April 2016 by the 1.8 m Ritchey-Chretien telescope of the Pan-STRARRS Project [10]. The current available orbital properties of this body were estimated through terrestrial observations spanning a data-arc of 5140 d or 14.07 yr, from 17th March of 2004 to 27th April of 2019. 2016 HO3 is classified as an Apollo asteroid moving in a low eccentricity (e = 0:10) and low inclination (I = 7:77). It orbits in a 1 : 1 mean motion resonance with Earth with a minimum orbital intersection distance of 0.03 AU. The physical parameters of 2016HO3 are currently poorly constrained. One single light curve from April 2017 gives a magnitude H = 24:3 and a rotation period = 0:467 ± 0:008 hr [11], suggesting it is a typical small fast rotator. The size of 2016 HO3 has not yet beenfirmly established, but it is likely about 40-100 m. Based on an assumed standard albedo for S-type asteroids of 0.20 and an absolute magnitude of 24.3, it measures 41 meters in diameter. A recent study by Reddy et al. [12] established a possible shape model suggesting that 2016 HO3 is likely an elongated asteroid, which has a b∕ a ratio smaller than 0.479. Color ratios derived from a low-resolution visible spectrum of 2016 HO3 suggest an S taxonomic type [11]. The particle size distribution and porosity of the regolith can be estimated if the thermal inertia or effective thermal conductivity of asteroids is known. However, the faintness of 2016 HO3 makes such an estimation almost impossible. Observations by the Hayabusa, Hayabusa 2, and OSIRISRex spacecraft revealed that the surface morphology of small-sized asteroids (less than 1 km) is mostly covered by boulders and in some areas by regolith, which is different from larger asteroids that are mostly covered byfinegrained regolith [3, 5, 13]. 2016 HO3 is a small asteroid with a diameter of 40-100 meters. It is reasonable to assume that it has a similar rubble pile surface. Therefore, for this study, we consider the surface of 2016 HO3 is covered by boulders with sizes ranging from several centimeters to at least several tens of centimeters andfine particles with sizes ranging from micrometer and millimeter. 3. Development of 2016 HO3 Simulants Minerals are the basic building blocks of planetary materials, and thus, mineral-based simulants offer the closest potential match to the properties of actual asteroid materials [7]. Both telescopic observations and analyses on the returned asteroid Itokawa sample indicate that most S-type bodies have mineralogy similar to those of ordinary chondrites [14, 15]. Ordinary chondrite meteorites (OCs) are silicate-rich meteorites and by far the most abundant meteorites (80% of all falls) [16]. According to their iron content, these meteorites are divided into H (high total Fe), L (low total Fe), and LL (low total Fe, low metal) groups, with 42.8%, 47.4%, and 9.8% of OC falls belonging to each group, respectively. Composition measurements of H, L, and LL group OCs provide a reference model from which to develop simulated asteroid 2016 HO3 materials for current observations suggests an S-type asteroid. The modal mineral abundance results obtained by X-ray diffraction analysis on 18 H, 17 L, and 13 LL unbrecciated OC falls show that olivine and low-Ca pyroxene are the most abundant phases present, while plagioclase, high-Ca pyroxene, troilite, and metal comprise the remaining XRD-measured mineralogy [17]. It is not clear that which OC group 2016 HO3 belongs to, and thus, we developed three prototype simulants QLS-1, QLS-2, and QLS-3 corresponding to H, L, and LL OCs, respectively. Table 1: Average modal abundances of ordinary chondrites (wt%, from [17]). Class H group L group LL group Number of samples 18 17 13 Olivine 33.0 42.1 51.1 Low-Ca pyroxene 25.6 23.2 21.1 High-Ca pyroxene 6.7 8.1 7.4 Plagioclase 9.0 9.4 9.7 Troilite 5.8 7.2 5.7 Metal 18.2 8.4 3.5 Othersa 1.7 1.6 1.6 a Normative abundances of apatite, ilmenite, and chromite [18]. Table 2: Mineral recipe for prototype simulant of 2016 HO3 (wt%). QLS-1 QLS-2 QLS-3 Olivine 30 38.7 46.9 Low-Ca pyroxene 29.1 28.2 25.6 High-Ca pyroxene 3.2 3.1 2.8 Plagioclase 9 9.4 9.7 Pyrite 5.8 7.2 5.7 Fe metal 19.4 10.6 7.2 Ni metal 1.8 0.8 0.3 Othersa 1.7 1.6 1.6 a Mixture of apatite, ilmenite, and chromite. 2 Space: Science & Technology Table 1 lists the average modal mineralogy of the aforementioned H, L, and LL OC falls. The mineral recipes for the prototype simulants are based on these data, with some modification because of various trades involved in sourcing raw materials (Table 2). To develop asteroid 2016 HO3 simulant, we obtained different types of raw rock or mineral materials from terrestrial mines, including olivine from Hannuoba area in Hebei Province, low-Ca pyroxene from Alashan area in Inner Mongolia Province, high-Ca pyroxene from Qingdao area in Shandong Province, and plagioclase from Shijiazhuang area in Hebei Province. Some raw materials are from commercial supplies, including powdery pyrite, metal iron and nickel, apatite, ilmenite, and chromite. Wefirst crushed the raw materials in chunk or other nonpowder forms intofine powders. Pyrite, metal iron, and nickel came from commercial suppliers as micronsized powders and require no further processing. The average grain size of crushed powder materials is 50-100 microns. After having all raw materials processed, we mixed the powder material in the appropriate weight ratios (Table 2), resulting in homogeneous medium to dark gray powders (Figure 1(a)). Due to the absence of natural binders (clays) in the OC composition, we added sodium metasilicate pentahydrate (a.k.a. water glass) into the mixture, which polymerizes into a rocky binder under modest heat and the“pentahydrate” is released during polymerization. We dissolved sodium metasilicate pentahydrate in pure water and then mixed this solution with the dry mixture (in a 1 : 5 ratio). The concentration of sodium metasilicate pentahydrate is 4-5% of the dry mixture. Then, the wet paste was dried in an oven under temperature at 150- 160° C. When the paste lost all its water and formed solid blocks, we crushed them using hand tools, such as a wooden club, which results in a power-law particle size distribution [19]. The size distribution could be adjusted by sieving and mixing. 4. Characteristics of 2016 HO3 Simulants We developed three prototype simulants with different abundance of iron and metal contents (QLS-1, QLS-2, and QLS-3), which accordingly represent asteroids linking to H, L, and LL type OCs (Figure 2). These asteroid simulant prototypes were characterized using scanning electron microscopy, X-rayfluorescence spectrometer, and VNIR reflectance spectroscopy. The three S-type asteroid prototype simulants were imaged using a Zeiss Super 55field emission scanning electron microscope (SEM) at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS), with 15 kV accelerating voltage. As shown in Figure 3, particles are subrounded to highly angular, with elongated, platy, and square shapes. Veryfine particles (<5μm) adhere to the surfaces of larger ones. Larger particles include individual minerals and agglomerate of smaller particles. The major element chemistry of the prototype simulants was measured and analyzed using XRF-1500 X-ray (a) (b) (c) (d) Figure 1: Production of 2016 HO3 prototype simulants. (a) Dry mixture of raw materials. (b) Paste after adding sodium metasilicate pentahydrate solution. (c) Driedfine and small-sized particles. (d) Large-sized boulders. Space: Science & Technology 3
第25页
15 for S-type asteroid are relatively less reported. A new Itokawa asteroid regolith simulant (called IRS-1) as an S-type asteroid was developed for China’s upcoming asteroid exploration [9]. However, this simulant only containsfine grains ranging from submillimeter to centimeter, without larger-sized boulders. This study designed and developed the morphology equivalent prototype simulants for 2016 HO3 that are with particle sizes ranging from micrometer to at least tens of centimeters. The goal of these simulants is to simulate a possible rubble pile morphology and the reflectance feature of the surface of 2016 HO3 so that they can serve practical uses during the mission preparation and operation phases, such as test and evaluation of navigation system and sampling device before the actual touchdown or landing operation. Furthermore, when the spacecraft’s initial observation reveals a different size distribution of 2016 HO3 surface than our estimation, which is always the case, our simulants can be easily modified to reflect the newfindings. 2. Observation of (469219) 2016 HO3 The near-earth asteroid (469219) 2016 HO3 was found on 27th of April 2016 by the 1.8 m Ritchey-Chretien telescope of the Pan-STRARRS Project [10]. The current available orbital properties of this body were estimated through terrestrial observations spanning a data-arc of 5140 d or 14.07 yr, from 17th March of 2004 to 27th April of 2019. 2016 HO3 is classified as an Apollo asteroid moving in a low eccentricity (e = 0:10) and low inclination (I = 7:77). It orbits in a 1 : 1 mean motion resonance with Earth with a minimum orbital intersection distance of 0.03 AU. The physical parameters of 2016HO3 are currently poorly constrained. One single light curve from April 2017 gives a magnitude H = 24:3 and a rotation period = 0:467 ± 0:008 hr [11], suggesting it is a typical small fast rotator. The size of 2016 HO3 has not yet beenfirmly established, but it is likely about 40-100 m. Based on an assumed standard albedo for S-type asteroids of 0.20 and an absolute magnitude of 24.3, it measures 41 meters in diameter. A recent study by Reddy et al. [12] established a possible shape model suggesting that 2016 HO3 is likely an elongated asteroid, which has a b∕ a ratio smaller than 0.479. Color ratios derived from a low-resolution visible spectrum of 2016 HO3 suggest an S taxonomic type [11]. The particle size distribution and porosity of the regolith can be estimated if the thermal inertia or effective thermal conductivity of asteroids is known. However, the faintness of 2016 HO3 makes such an estimation almost impossible. Observations by the Hayabusa, Hayabusa 2, and OSIRISRex spacecraft revealed that the surface morphology of small-sized asteroids (less than 1 km) is mostly covered by boulders and in some areas by regolith, which is different from larger asteroids that are mostly covered byfinegrained regolith [3, 5, 13]. 2016 HO3 is a small asteroid with a diameter of 40-100 meters. It is reasonable to assume that it has a similar rubble pile surface. Therefore, for this study, we consider the surface of 2016 HO3 is covered by boulders with sizes ranging from several centimeters to at least several tens of centimeters andfine particles with sizes ranging from micrometer and millimeter. 3. Development of 2016 HO3 Simulants Minerals are the basic building blocks of planetary materials, and thus, mineral-based simulants offer the closest potential match to the properties of actual asteroid materials [7]. Both telescopic observations and analyses on the returned asteroid Itokawa sample indicate that most S-type bodies have mineralogy similar to those of ordinary chondrites [14, 15]. Ordinary chondrite meteorites (OCs) are silicate-rich meteorites and by far the most abundant meteorites (80% of all falls) [16]. According to their iron content, these meteorites are divided into H (high total Fe), L (low total Fe), and LL (low total Fe, low metal) groups, with 42.8%, 47.4%, and 9.8% of OC falls belonging to each group, respectively. Composition measurements of H, L, and LL group OCs provide a reference model from which to develop simulated asteroid 2016 HO3 materials for current observations suggests an S-type asteroid. The modal mineral abundance results obtained by X-ray diffraction analysis on 18 H, 17 L, and 13 LL unbrecciated OC falls show that olivine and low-Ca pyroxene are the most abundant phases present, while plagioclase, high-Ca pyroxene, troilite, and metal comprise the remaining XRD-measured mineralogy [17]. It is not clear that which OC group 2016 HO3 belongs to, and thus, we developed three prototype simulants QLS-1, QLS-2, and QLS-3 corresponding to H, L, and LL OCs, respectively. Table 1: Average modal abundances of ordinary chondrites (wt%, from [17]). Class H group L group LL group Number of samples 18 17 13 Olivine 33.0 42.1 51.1 Low-Ca pyroxene 25.6 23.2 21.1 High-Ca pyroxene 6.7 8.1 7.4 Plagioclase 9.0 9.4 9.7 Troilite 5.8 7.2 5.7 Metal 18.2 8.4 3.5 Othersa 1.7 1.6 1.6 a Normative abundances of apatite, ilmenite, and chromite [18]. Table 2: Mineral recipe for prototype simulant of 2016 HO3 (wt%). QLS-1 QLS-2 QLS-3 Olivine 30 38.7 46.9 Low-Ca pyroxene 29.1 28.2 25.6 High-Ca pyroxene 3.2 3.1 2.8 Plagioclase 9 9.4 9.7 Pyrite 5.8 7.2 5.7 Fe metal 19.4 10.6 7.2 Ni metal 1.8 0.8 0.3 Othersa 1.7 1.6 1.6 a Mixture of apatite, ilmenite, and chromite. 2 Space: Science & Technology Table 1 lists the average modal mineralogy of the aforementioned H, L, and LL OC falls. The mineral recipes for the prototype simulants are based on these data, with some modification because of various trades involved in sourcing raw materials (Table 2). To develop asteroid 2016 HO3 simulant, we obtained different types of raw rock or mineral materials from terrestrial mines, including olivine from Hannuoba area in Hebei Province, low-Ca pyroxene from Alashan area in Inner Mongolia Province, high-Ca pyroxene from Qingdao area in Shandong Province, and plagioclase from Shijiazhuang area in Hebei Province. Some raw materials are from commercial supplies, including powdery pyrite, metal iron and nickel, apatite, ilmenite, and chromite. Wefirst crushed the raw materials in chunk or other nonpowder forms intofine powders. Pyrite, metal iron, and nickel came from commercial suppliers as micronsized powders and require no further processing. The average grain size of crushed powder materials is 50-100 microns. After having all raw materials processed, we mixed the powder material in the appropriate weight ratios (Table 2), resulting in homogeneous medium to dark gray powders (Figure 1(a)). Due to the absence of natural binders (clays) in the OC composition, we added sodium metasilicate pentahydrate (a.k.a. water glass) into the mixture, which polymerizes into a rocky binder under modest heat and the“pentahydrate” is released during polymerization. We dissolved sodium metasilicate pentahydrate in pure water and then mixed this solution with the dry mixture (in a 1 : 5 ratio). The concentration of sodium metasilicate pentahydrate is 4-5% of the dry mixture. Then, the wet paste was dried in an oven under temperature at 150- 160° C. When the paste lost all its water and formed solid blocks, we crushed them using hand tools, such as a wooden club, which results in a power-law particle size distribution [19]. The size distribution could be adjusted by sieving and mixing. 4. Characteristics of 2016 HO3 Simulants We developed three prototype simulants with different abundance of iron and metal contents (QLS-1, QLS-2, and QLS-3), which accordingly represent asteroids linking to H, L, and LL type OCs (Figure 2). These asteroid simulant prototypes were characterized using scanning electron microscopy, X-rayfluorescence spectrometer, and VNIR reflectance spectroscopy. The three S-type asteroid prototype simulants were imaged using a Zeiss Super 55field emission scanning electron microscope (SEM) at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS), with 15 kV accelerating voltage. As shown in Figure 3, particles are subrounded to highly angular, with elongated, platy, and square shapes. Veryfine particles (<5μm) adhere to the surfaces of larger ones. Larger particles include individual minerals and agglomerate of smaller particles. The major element chemistry of the prototype simulants was measured and analyzed using XRF-1500 X-ray (a) (b) (c) (d) Figure 1: Production of 2016 HO3 prototype simulants. (a) Dry mixture of raw materials. (b) Paste after adding sodium metasilicate pentahydrate solution. (c) Driedfine and small-sized particles. (d) Large-sized boulders. Space: Science & Technology 3
第26页
16 fluorescence (XRF) with RSD between 0.1 and 1% at IGGCAS, and the results are listed in Table 3. Compared to H, L, and LL OCs, the simulants have significantly lower TFe2O3 (Table 3), a consequence of using terrestrial low-Fe pyroxene and olivine as the raw material. Reflectance spectra of the prototype simulants were acquired at the China University of Geosciences (Wuhan), using a VERTEX 70v Fourier transform infrared spectrometer at ambient condition. The scattering configuration was approximately 45° -45° biconical. The resulting spectra (0.500–2.5μm) are shown in Figure 4, which offsets them for clarity. As shown in Figure 4, the measured spectrum of the three prototype simulants is broadly similar to the spectrum for asteroid Itokawa and S-type asteroids. Both samples have two strong absorption bands at ~0.9–1μm (a) (b) (c) (d) Figure 2: Scanning electron microscope images of typical particles of prototype simulants of 2016 HO3. (a) Overview of H group prototype simulant QLS-1 particles in various sizes and shapes. The inset white square indicates where the image (b) is taken. (b) Zoom in of image (a) showing mixture of large andfine grains. (c) LL group prototype simulant QLS-3. (d) Fines coating the surface of a larger grain in the L group prototype simulant. 0.5 0.9 1.0 1.1 1.2 1.3 1.4 1.0 1.5 Wavelength (m) Normalized reflectance 2.0 QLS-1 QLS-2 QLS-3 S-type asteroid 25143 ltokawa 2.5 Figure 3: VNIR reflectance spectra of 2016 HO3 prototype simulants compared with asteroid 25143 Itokawa and S-type asteroids [21]. Table 3: Major element bulk chemistry of prototype simulants and ordinary chondritesa . Oxide QLS-1 H group QLS-2 L group QLS-3 LL group SiO2 34.4 34.2 36.6 37.9 38.7 39.7 TiO2 0.2 0.1 0.2 0.1 0.2 0.1 Al2O3 4.4 1.9 4.3 2.1 4.3 2.2 TFe2O3 27.6 35.4 22.0 29.5 18.7 26.8 MnO 0.2 0.3 0.2 0.3 0.2 0.4 MgO 20.8 21.8 24.1 23.8 26.9 24.8 CaO 5.6 1.6 5.6 1.7 5.7 1.8 Na2O 0.7 0.7 0.7 0.9 0.6 0.9 K2O 0.2 0.1 0.2 0.1 0.2 0.1 P2O5 0.4 0.2 0.3 0.2 0.4 0.2 NiO 2.4 2.0 1.2 1.6 0.6 1.4 Cr2O3 0.9 0.5 1.0 0.6 1.0 0.5 a From [20]. 4 Space: Science & Technology and ~1.9μm. The ~0.9–1μm mineral absorption band is the result of the presence of either pyroxene or olivine or both, while the ~1.9μm band is indicative of the presence of pyroxene [21]. Weak absorption bands at ~1.4μm are related to the OH stretching associated with structurally bound OH and H2O, while the ~2.2μm and ~2.3μm absorptions may be due to metal-OH (e.g., Al-OH and Fe, Mg-OH) transitions, which is likely caused by vapor in the ambient environment and/or impurities in the raw materials. 5. Application Our primary purpose for developing 2016 HO3 simulants is to reproduce the optical properties and a possible rubble pile structure of the surface. Since the morphology of 2016 HO3 is still unknown, asteroid Itokawa was chosen here as a reference. We sieved and mixed large boulders (several to several tens of centimeters) and smaller particles to match the size distributions of particles roughly following the powerlaw index as measured on Itokawa [4, 22]. The simulants were distributed in a Styrofoam container with the size of 1 m∗ 0:8 m. Comparison between images of our simulated topography and Itokawa surface demonstrates similar morphological characteristics (Figure 4). When the optical properties of 2016 HO3 are revealed by the initial observations of the spacecraft during the phase mission, our prototype simulants could be easily adjusted to match possible new and unexpectedfindings. Subsequently, the simulants could be used in optical navigation experiment and to study the efficiency of the sampler before its touchdown or landing. 6. Conclusion We used remote observations of asteroid 2016 HO3 and modal mineralogy data from ordinary chondrites to develop three prototype simulants of S-type asteroids for engineering evaluation of the optical navigation system and the sampling device of the spacecraft. The size of the simulant particles ranges from tens of centimeter to micrometer. The characterizing results of the prototype simulants show similarity to their analogue meteorites despite some discrepancies in bulk chemistry and reflectance spectra due to unavoidable factors in choosing terrestrial raw materials. These prototype simulants are easy to make and modify according to need. Data Availability The data used to support thefindings of this study are included within the article. Conflicts of Interest The authors declare that there is no conflict of interest regarding the publication of this article. Authors’ Contributions Yuechen Luo and Yuan Xiao contributed equally to this work. Acknowledgments We thank Dr. Danping Zhang, Dr. Lixin Gu, Prof. Jinhua Li, Dr. Te Jiang, and Prof. Hao Zhang for their assistance with the characterizing experiments and discussion in interpreting data. This work was supported by the Civil Aerospace Pre-research Project (D020302) and the Qian Xuesen Laboratory Innovation Funds (Y-KC-WY-99-ZC-005). References [1] P. Metzger, D. Britt, S. Covey, and J. S. Lewis, Figure of merit for asteroid regolith simulants, vol. 11, article EPSC2017-436, European Planetary Science Congress, 2017. [2] H. Miyamoto and T. Niihara,“Simplified simulated materials of asteroid Ryugu for spacecraft operations and scientific evaluations,” Natural Resources Research, vol. 30, no. 4, pp. 3035– 3044, 2021. [3] T. Michikami, C. Honda, H. Miyamoto et al.,“Boulder size and shape distributions on asteroid Ryugu,” Icarus, vol. 331, pp. 179–191, 2019. (a) 10 cm (b) 1 m Figure 4: (a) The simulated rough surface using 2016 HO3 simulant and (b) the close-up image of the very rough surface of Itokawa [23]. Space: Science & Technology 5
第27页
17 fluorescence (XRF) with RSD between 0.1 and 1% at IGGCAS, and the results are listed in Table 3. Compared to H, L, and LL OCs, the simulants have significantly lower TFe2O3 (Table 3), a consequence of using terrestrial low-Fe pyroxene and olivine as the raw material. Reflectance spectra of the prototype simulants were acquired at the China University of Geosciences (Wuhan), using a VERTEX 70v Fourier transform infrared spectrometer at ambient condition. The scattering configuration was approximately 45° -45° biconical. The resulting spectra (0.500–2.5μm) are shown in Figure 4, which offsets them for clarity. As shown in Figure 4, the measured spectrum of the three prototype simulants is broadly similar to the spectrum for asteroid Itokawa and S-type asteroids. Both samples have two strong absorption bands at ~0.9–1μm (a) (b) (c) (d) Figure 2: Scanning electron microscope images of typical particles of prototype simulants of 2016 HO3. (a) Overview of H group prototype simulant QLS-1 particles in various sizes and shapes. The inset white square indicates where the image (b) is taken. (b) Zoom in of image (a) showing mixture of large andfine grains. (c) LL group prototype simulant QLS-3. (d) Fines coating the surface of a larger grain in the L group prototype simulant. 0.5 0.9 1.0 1.1 1.2 1.3 1.4 1.0 1.5 Wavelength (m) Normalized reflectance 2.0 QLS-1 QLS-2 QLS-3 S-type asteroid 25143 ltokawa 2.5 Figure 3: VNIR reflectance spectra of 2016 HO3 prototype simulants compared with asteroid 25143 Itokawa and S-type asteroids [21]. Table 3: Major element bulk chemistry of prototype simulants and ordinary chondritesa . Oxide QLS-1 H group QLS-2 L group QLS-3 LL group SiO2 34.4 34.2 36.6 37.9 38.7 39.7 TiO2 0.2 0.1 0.2 0.1 0.2 0.1 Al2O3 4.4 1.9 4.3 2.1 4.3 2.2 TFe2O3 27.6 35.4 22.0 29.5 18.7 26.8 MnO 0.2 0.3 0.2 0.3 0.2 0.4 MgO 20.8 21.8 24.1 23.8 26.9 24.8 CaO 5.6 1.6 5.6 1.7 5.7 1.8 Na2O 0.7 0.7 0.7 0.9 0.6 0.9 K2O 0.2 0.1 0.2 0.1 0.2 0.1 P2O5 0.4 0.2 0.3 0.2 0.4 0.2 NiO 2.4 2.0 1.2 1.6 0.6 1.4 Cr2O3 0.9 0.5 1.0 0.6 1.0 0.5 a From [20]. 4 Space: Science & Technology and ~1.9μm. The ~0.9–1μm mineral absorption band is the result of the presence of either pyroxene or olivine or both, while the ~1.9μm band is indicative of the presence of pyroxene [21]. Weak absorption bands at ~1.4μm are related to the OH stretching associated with structurally bound OH and H2O, while the ~2.2μm and ~2.3μm absorptions may be due to metal-OH (e.g., Al-OH and Fe, Mg-OH) transitions, which is likely caused by vapor in the ambient environment and/or impurities in the raw materials. 5. Application Our primary purpose for developing 2016 HO3 simulants is to reproduce the optical properties and a possible rubble pile structure of the surface. Since the morphology of 2016 HO3 is still unknown, asteroid Itokawa was chosen here as a reference. We sieved and mixed large boulders (several to several tens of centimeters) and smaller particles to match the size distributions of particles roughly following the powerlaw index as measured on Itokawa [4, 22]. The simulants were distributed in a Styrofoam container with the size of 1 m∗ 0:8 m. Comparison between images of our simulated topography and Itokawa surface demonstrates similar morphological characteristics (Figure 4). When the optical properties of 2016 HO3 are revealed by the initial observations of the spacecraft during the phase mission, our prototype simulants could be easily adjusted to match possible new and unexpectedfindings. Subsequently, the simulants could be used in optical navigation experiment and to study the efficiency of the sampler before its touchdown or landing. 6. Conclusion We used remote observations of asteroid 2016 HO3 and modal mineralogy data from ordinary chondrites to develop three prototype simulants of S-type asteroids for engineering evaluation of the optical navigation system and the sampling device of the spacecraft. The size of the simulant particles ranges from tens of centimeter to micrometer. The characterizing results of the prototype simulants show similarity to their analogue meteorites despite some discrepancies in bulk chemistry and reflectance spectra due to unavoidable factors in choosing terrestrial raw materials. These prototype simulants are easy to make and modify according to need. Data Availability The data used to support thefindings of this study are included within the article. Conflicts of Interest The authors declare that there is no conflict of interest regarding the publication of this article. Authors’ Contributions Yuechen Luo and Yuan Xiao contributed equally to this work. Acknowledgments We thank Dr. Danping Zhang, Dr. Lixin Gu, Prof. Jinhua Li, Dr. Te Jiang, and Prof. Hao Zhang for their assistance with the characterizing experiments and discussion in interpreting data. This work was supported by the Civil Aerospace Pre-research Project (D020302) and the Qian Xuesen Laboratory Innovation Funds (Y-KC-WY-99-ZC-005). References [1] P. Metzger, D. Britt, S. Covey, and J. S. Lewis, Figure of merit for asteroid regolith simulants, vol. 11, article EPSC2017-436, European Planetary Science Congress, 2017. [2] H. Miyamoto and T. Niihara,“Simplified simulated materials of asteroid Ryugu for spacecraft operations and scientific evaluations,” Natural Resources Research, vol. 30, no. 4, pp. 3035– 3044, 2021. [3] T. Michikami, C. Honda, H. Miyamoto et al.,“Boulder size and shape distributions on asteroid Ryugu,” Icarus, vol. 331, pp. 179–191, 2019. (a) 10 cm (b) 1 m Figure 4: (a) The simulated rough surface using 2016 HO3 simulant and (b) the close-up image of the very rough surface of Itokawa [23]. Space: Science & Technology 5
第28页
18 [4] T. Michikami, A. M. Nakamura, N. Hirata et al.,“Size-frequency statistics of boulders on global surface of asteroid 25143 Itokawa,” Earth, Planets and Space, vol. 60, no. 1, pp. 13–20, 2008. [5] The OSIRIS-REx Team, K. J. Walsh, E. R. Jawin et al.,“Craters, boulders and regolith of (101955) Bennu indicative of an old and dynamic surface,” Nature Geoscience, vol. 12, no. 4, pp. 242–246, 2019. [6] J. Huang, X. Zhang, T. Wang, Z. Huo, X. Shi, and L. Meng, “Small body exploration in China,” European Planetary Science Congress, 2020. [7] D. T. Britt, K. M. Cannon, K. Donaldson Hanna et al.,“Simulated asteroid materials based on carbonaceous chondrite mineralogies,” Meteoritics and Planetary Science, vol. 54, no. 9, pp. 2067–2082, 2019. [8] A. R. Hildebrand, L. T. J. Hanton, M. Rankin, and M. I. Ibrahim,“An asteroid regolith simulant for hydrated carbonaceous chondrite lithologies (HCCL-1),” 78th Annual Meeting of the Meteoritical Society, vol. 1856, article 5368, 2015. [9] X. Zeng, X. Li, D. J. P. Martin et al.,“The Itokawa regolith simulant IRS-1 as an S-type asteroid surface analogue,” Icarus, vol. 333, pp. 371–384, 2019. [10] C. de la Fuente Marcos and R. de la Fuente Marcos,“Asteroid (469219) 2016 HO3, the smallest and closest Earth quasi-satellite,” Monthly Notices of the Royal Astronomical Society, vol. 462, no. 4, pp. 3441–3456, 2016. [11] X. Li and D. J. Scheeres,“The shape and surface environment of 2016 HO3,” Icarus, vol. 357, p. 114249, 2021. [12] V. Reddy, O. Kuhn, A. Thirouin et al.,“Ground-based characterization of Earth quasi satellite (469219) 2016 HO3,” AAS/- Division for Planetary Sciences Meeting Abstracts# 49, vol. 49, 2017August 2021. https://ui.adsabs.harvard.edu/abs/2017DPS ....4920407R. [13] O. S. Barnouin-Jha, A. F. Cheng, T. Mukai et al.,“Small-scale topography of 25143 Itokawa from the Hayabusa laser altimeter,” Icarus, vol. 198, no. 1, pp. 108–124, 2008. [14] T. Nakamura, T. Noguchi, M. Tanaka et al.,“Itokawa dust particles: a direct link between S-type asteroids and ordinary chondrites,” Science, vol. 333, no. 6046, pp. 1113–1116, 2011. [15] E. A. Cloutis, R. P. Binzel, and M. J. Gaffey,“Establishing asteroid-meteorite links,” Elements, vol. 10, no. 1, pp. 25–30, 2014. [16] R. Hutchison,“Meteorites: A Petrologic,” in Chemical and Isotopic Synthesis, Cambridge University Press, 2006. [17] T. L. Dunn, T. J. McCoy, J. M. Sunshine, and H. Y. McSween Jr.,“A coordinated spectral, mineralogical, and compositional study of ordinary chondrites,” Icarus, vol. 208, no. 2, pp. 789– 797, 2010. [18] H. Y. McSween, M. E. Bennett, and E. Jarosewich,“The mineralogy of ordinary chondrites and implications for asteroid spectrophotometry,” Icarus, vol. 90, no. 1, pp. 107–116, 1991. [19] D. L. Turcotte,“Fractals and fragmentation,” Journal of Geophysical Research: Solid Earth, vol. 91, no. B2, p. 1921, 1986. [20] C. R. Fulton and J. M. Rhodes,“The chemistry and origin of the ordinary chondrites: implications from refractorylithophile and siderophile elements,” Journal of Geophysical Research: Solid Earth, vol. 89, no. S02, article B543, 1984. [21] R. P. Binzel, A. S. Rivkin, S. J. Bus, J. M. Sunshine, and T. H. Burbine,“MUSES-C target asteroid (25143) 1998 SF36: a reddened ordinary chondrite,” Meteoritics and Planetary Science, vol. 36, no. 8, pp. 1167–1172, 2001. [22] T. Michikami and A. Hagermann,“Boulder sizes and shapes on asteroids: a comparative study of Eros, Itokawa and Ryugu,” Icarus, vol. 357, p. 114282, 2021. [23] P. Michel,“Formation and physical properties of asteroids,” Elements, vol. 10, no. 1, pp. 19–24, 2014. 6 Space: Science & Technology Research Article Performance Evaluation Indicators of Space Dynamic Networks under Broadcast Mechanism Zipeng Ye and Qingrui Zhou Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100094, China Correspondence should be addressed to Qingrui Zhou; zhouqingrui@qxslab.cn Received 6 August 2021; Accepted 17 October 2021; Published 29 October 2021 Copyright © 2021 Zipeng Ye and Qingrui Zhou. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). Large-scale heterogeneous constellations will be the major forms of future space-based systems, and the implementation of numerous derived applications depends mainly on intersatellite communication. The nodes representing heterogeneous satellites will form the networks with rapidly changing topology. However, few researches have been carried out for such networks. This paper studies the broadcast mechanism for space dynamic networks and establishes centralized and distributed routing framework. And then, performance evaluation indicators are proposed to evaluate both the connectivity of dynamic networks and the effectiveness of routing algorithms. Finally, we examine the performance of multigroup networks and verify the rationality of corresponding indicators. We also explore the impact of information survival time which directly affects the delivery ratio and, if unfortunately, may waste the communication resources. Empirical conclusion about the survival time is given in thefinal part. We believe the performance indicators and the routing algorithms proposed in this paper are great help to future space-based system and both the broadcast mechanism designing. 1. Introduction With the development of space technology, the cost of deploying low-earth orbit (LEO) spacecrafts is rapidly decreasing. In recent years, a series of LEO satellite cluster programs have been proposed by several institutes [1–3]. The most representative one is the Blackjack Pit Boss program proposed by Defense Advanced Research Projects Agency (DARPA). Through communicating at the commercial data transmission layer and combining with onboard data processing software, it can realize various strategic missions such as target identifying, tracking, surveilling, area reconnoitering, and all-weather multidomain resource detecting without the involvement of human. Data sharing by effective communication mechanism is a prerequisite for these cluster programs to accomplish data fusion, autonomous decision-making, and collaboration. Therefore, it is necessary to make a profound study on the network structure and communication mechanism of satellite cluster. The delay/disruption tolerant network (DTN) model proposed by Intel and NASA has been widely used for space networks [4, 5]. The Consultative Committee for Space Data Systems (CCSDS) has also built a DTN working group to promote the standardization of DTN technology. DTN adopts“store-wait-forward” routing mechanism, that is, an intermediate node stores information temporarily, waits for the establishment of a communication link with the next hop node, and then forwards the information. By repeating this mechanism, information is transmitted from the source node to the destination node. The most well-known routing algorithm for the DTN model is the epidemic algorithm [6]. It can reach the highest delivery ratio in the shortest period without any prior knowledge assistance. However, epidemic algorithm leads to a severe wasting of resources. An improved and universal method called PRoPHET can efficiently cut down the overhead by some constrains [7], that is, nodes deliver messages through the similarity which decided by the history, i.e., the implement of the PRoPHET algorithm requires prior knowledge. There are other routing algorithms proposed in [8–11], both of whom areflooding-based and can achieve trade-off through dropping or reducing policies. Indeed, all these algorithms can realize high delivery ratio and low average delay if the resources are unlimited. Based on the predictability of space networks, Merugu and Zegura and Fischer et al. have studied the routing AAAS Space: Science & Technology Volume 2021, Article ID 9826517, 11 pages https://doi.org/10.34133/2021/9826517
第29页
19 [4] T. Michikami, A. M. Nakamura, N. Hirata et al.,“Size-frequency statistics of boulders on global surface of asteroid 25143 Itokawa,” Earth, Planets and Space, vol. 60, no. 1, pp. 13–20, 2008. [5] The OSIRIS-REx Team, K. J. Walsh, E. R. Jawin et al.,“Craters, boulders and regolith of (101955) Bennu indicative of an old and dynamic surface,” Nature Geoscience, vol. 12, no. 4, pp. 242–246, 2019. [6] J. Huang, X. Zhang, T. Wang, Z. Huo, X. Shi, and L. Meng, “Small body exploration in China,” European Planetary Science Congress, 2020. [7] D. T. Britt, K. M. Cannon, K. Donaldson Hanna et al.,“Simulated asteroid materials based on carbonaceous chondrite mineralogies,” Meteoritics and Planetary Science, vol. 54, no. 9, pp. 2067–2082, 2019. [8] A. R. Hildebrand, L. T. J. Hanton, M. Rankin, and M. I. Ibrahim,“An asteroid regolith simulant for hydrated carbonaceous chondrite lithologies (HCCL-1),” 78th Annual Meeting of the Meteoritical Society, vol. 1856, article 5368, 2015. [9] X. Zeng, X. Li, D. J. P. Martin et al.,“The Itokawa regolith simulant IRS-1 as an S-type asteroid surface analogue,” Icarus, vol. 333, pp. 371–384, 2019. [10] C. de la Fuente Marcos and R. de la Fuente Marcos,“Asteroid (469219) 2016 HO3, the smallest and closest Earth quasi-satellite,” Monthly Notices of the Royal Astronomical Society, vol. 462, no. 4, pp. 3441–3456, 2016. [11] X. Li and D. J. Scheeres,“The shape and surface environment of 2016 HO3,” Icarus, vol. 357, p. 114249, 2021. [12] V. Reddy, O. Kuhn, A. Thirouin et al.,“Ground-based characterization of Earth quasi satellite (469219) 2016 HO3,” AAS/- Division for Planetary Sciences Meeting Abstracts# 49, vol. 49, 2017August 2021. https://ui.adsabs.harvard.edu/abs/2017DPS ....4920407R. [13] O. S. Barnouin-Jha, A. F. Cheng, T. Mukai et al.,“Small-scale topography of 25143 Itokawa from the Hayabusa laser altimeter,” Icarus, vol. 198, no. 1, pp. 108–124, 2008. [14] T. Nakamura, T. Noguchi, M. Tanaka et al.,“Itokawa dust particles: a direct link between S-type asteroids and ordinary chondrites,” Science, vol. 333, no. 6046, pp. 1113–1116, 2011. [15] E. A. Cloutis, R. P. Binzel, and M. J. Gaffey,“Establishing asteroid-meteorite links,” Elements, vol. 10, no. 1, pp. 25–30, 2014. [16] R. Hutchison,“Meteorites: A Petrologic,” in Chemical and Isotopic Synthesis, Cambridge University Press, 2006. [17] T. L. Dunn, T. J. McCoy, J. M. Sunshine, and H. Y. McSween Jr.,“A coordinated spectral, mineralogical, and compositional study of ordinary chondrites,” Icarus, vol. 208, no. 2, pp. 789– 797, 2010. [18] H. Y. McSween, M. E. Bennett, and E. Jarosewich,“The mineralogy of ordinary chondrites and implications for asteroid spectrophotometry,” Icarus, vol. 90, no. 1, pp. 107–116, 1991. [19] D. L. Turcotte,“Fractals and fragmentation,” Journal of Geophysical Research: Solid Earth, vol. 91, no. B2, p. 1921, 1986. [20] C. R. Fulton and J. M. Rhodes,“The chemistry and origin of the ordinary chondrites: implications from refractorylithophile and siderophile elements,” Journal of Geophysical Research: Solid Earth, vol. 89, no. S02, article B543, 1984. [21] R. P. Binzel, A. S. Rivkin, S. J. Bus, J. M. Sunshine, and T. H. Burbine,“MUSES-C target asteroid (25143) 1998 SF36: a reddened ordinary chondrite,” Meteoritics and Planetary Science, vol. 36, no. 8, pp. 1167–1172, 2001. [22] T. Michikami and A. Hagermann,“Boulder sizes and shapes on asteroids: a comparative study of Eros, Itokawa and Ryugu,” Icarus, vol. 357, p. 114282, 2021. [23] P. Michel,“Formation and physical properties of asteroids,” Elements, vol. 10, no. 1, pp. 19–24, 2014. 6 Space: Science & Technology Research Article Performance Evaluation Indicators of Space Dynamic Networks under Broadcast Mechanism Zipeng Ye and Qingrui Zhou Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100094, China Correspondence should be addressed to Qingrui Zhou; zhouqingrui@qxslab.cn Received 6 August 2021; Accepted 17 October 2021; Published 29 October 2021 Copyright © 2021 Zipeng Ye and Qingrui Zhou. Exclusive Licensee Beijing Institute of Technology Press. Distributed under a Creative Commons Attribution License (CC BY 4.0). Large-scale heterogeneous constellations will be the major forms of future space-based systems, and the implementation of numerous derived applications depends mainly on intersatellite communication. The nodes representing heterogeneous satellites will form the networks with rapidly changing topology. However, few researches have been carried out for such networks. This paper studies the broadcast mechanism for space dynamic networks and establishes centralized and distributed routing framework. And then, performance evaluation indicators are proposed to evaluate both the connectivity of dynamic networks and the effectiveness of routing algorithms. Finally, we examine the performance of multigroup networks and verify the rationality of corresponding indicators. We also explore the impact of information survival time which directly affects the delivery ratio and, if unfortunately, may waste the communication resources. Empirical conclusion about the survival time is given in thefinal part. We believe the performance indicators and the routing algorithms proposed in this paper are great help to future space-based system and both the broadcast mechanism designing. 1. Introduction With the development of space technology, the cost of deploying low-earth orbit (LEO) spacecrafts is rapidly decreasing. In recent years, a series of LEO satellite cluster programs have been proposed by several institutes [1–3]. The most representative one is the Blackjack Pit Boss program proposed by Defense Advanced Research Projects Agency (DARPA). Through communicating at the commercial data transmission layer and combining with onboard data processing software, it can realize various strategic missions such as target identifying, tracking, surveilling, area reconnoitering, and all-weather multidomain resource detecting without the involvement of human. Data sharing by effective communication mechanism is a prerequisite for these cluster programs to accomplish data fusion, autonomous decision-making, and collaboration. Therefore, it is necessary to make a profound study on the network structure and communication mechanism of satellite cluster. The delay/disruption tolerant network (DTN) model proposed by Intel and NASA has been widely used for space networks [4, 5]. The Consultative Committee for Space Data Systems (CCSDS) has also built a DTN working group to promote the standardization of DTN technology. DTN adopts“store-wait-forward” routing mechanism, that is, an intermediate node stores information temporarily, waits for the establishment of a communication link with the next hop node, and then forwards the information. By repeating this mechanism, information is transmitted from the source node to the destination node. The most well-known routing algorithm for the DTN model is the epidemic algorithm [6]. It can reach the highest delivery ratio in the shortest period without any prior knowledge assistance. However, epidemic algorithm leads to a severe wasting of resources. An improved and universal method called PRoPHET can efficiently cut down the overhead by some constrains [7], that is, nodes deliver messages through the similarity which decided by the history, i.e., the implement of the PRoPHET algorithm requires prior knowledge. There are other routing algorithms proposed in [8–11], both of whom areflooding-based and can achieve trade-off through dropping or reducing policies. Indeed, all these algorithms can realize high delivery ratio and low average delay if the resources are unlimited. Based on the predictability of space networks, Merugu and Zegura and Fischer et al. have studied the routing AAAS Space: Science & Technology Volume 2021, Article ID 9826517, 11 pages https://doi.org/10.34133/2021/9826517 Performance Evaluation Indicators of Space Dynamic Networks under Broadcast Mechanism Zipeng Ye and Qingrui Zhou Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology, Beijing 100094, China Correspondence should be addressed to Qingrui Zhou; zhouqingrui@qxslab.cn Abstract: Large-scale heterogeneous constellations will be the major forms of future space-based systems, and the implementation of numerous derived applications depends mainly on intersatellite communication. The nodes representing heterogeneous satellites will form the networks with rapidly changing topology. However, few researches have been carried out for such networks. This paper studies the broadcast mechanism for space dynamic networks and establishes centralized and distributed routing framework. And then, performance evaluation indicators are proposed to evaluate both the connectivity of dynamic networks and the effectiveness of routing algorithms. Finally, we examine the performance of multigroup networks and verify the rationality of corresponding indicators. We also explore the impact of information survival time which directly affects the delivery ratio and, if unfortunately, may waste the communication resources. Empirical conclusion about the survival time is given in the final part. We believe the performance indicators and the routing algorithms proposed in this paper are great help to future space-based system and both the broadcast mechanism designing.
第30页
20 framework, routing algorithms, and the effective solutions to unexpected situation [12, 13]. Their main works include designing both the space-time graph and the optimal transmission routing table from the source node to the destination node. Such space-time graph-based algorithms are channel resources friendly, yet they have high computational complexity. Similar space-time graph-based algorithms can be referred to [14–17]. Yan et al. and Wang et al. have evaluated these typical DTN routing algorithms and designed an onboard routing algorithm [18, 19]. Based on the transmission delay, delivery ratio, they have given some suggestions for future space DTN research. Scholars have conducted extensive research on space DTN, and most of these focus on end-to-end information transmission. Future LEO satellite clusters, such like Blackjack program, are more dependent on information sharing among all satellites, so as to achieve efficient data fusion, autonomous decision-making, and collaboration. However, there is still a lack of research on space dynamic networks under this broadcasting mechanism. This paper is aimed at proposing some metrics that can effectively evaluate the network performance and routing algorithm performance under broadcasting mechanism and providing valuable ideas for subsequent research. 2. Model Static network has simpler model, with more complete research in theory and engineering implementations. The dynamic network is totally different since the dynamic network is with sequence and irreversibility. For example, if node A is connecting with node B and B is connecting to C, then there is a link between A and C. However, in dynamic network, if the connection between A and B occurs after B and C, then we cannot assert that there is a link between A and C. Dynamic networks are usually divided into two forms: lossless form and lossy form. The lossless form records all the connection status, while lossy form divides timespan into several segments, regards every segment as a static network, andfinally connects all segments to form a complete network. The lossless form needs to record every moment that connection relationship has changed, which requires a large amount of storage. Therefore, the lossy form can be used to model a dynamic network [12, 13]. Considering a quaternion, c = ð Þ a, b, t,δt , ð1Þ where fa, bg represents the two nodes, respectively, and denotes the moment connection occurring andδt denotes the duration of the connection. For a segment ½tm, tm +Δt�, if c satisfies one of the following conditions, tm≤ t < tm +Δt, tm≤ t +δt < tm +Δt, t≤ tm < tm +Δt < t +δt: ð2Þ Meeting any of the above conditions means that there is always a part of time in the whole connecting duration that is within the specified segment ½tm, tm +Δt�. Then, an edge is created between the nodes, and all such edges form the set of edges Em. Let the snapshot of segment ½tm, tm +Δt� be Sm = ðN, EmÞ. Then, we can model the dynamic network as G = ðI, SÞ, where I represents the total timespan, I = ½Tstart, Tend�, and S is the set of all snapshots. Accumulating each small segment can obtain the entire networks. As shown in Figure 1, there exist four nodes N = fA, B, C, Dg, and the subgraphs of Figure 1 are snapshots fS1, S2, S3, S4g. At time t1, if node A needs to transmit information to target node C, there are two routing methods: (1) information is transmitted from node A to node D at time t1 and then transmitted from node D to node C at t2, and this “store-wait-forward” routing way belongs to the DTN method; (2) node A waits until t3 to establish a complete path with node C, and this routing method is often used for ad hoc networks. Obviously, the complete path between the source and the destination may not always appear, so the DTN routing method can not only realize the information delivery in a shorter time, but also improve the delivery ratio [20–22]. 3. Routing in Space Networks under Broadcasting Mechanism Traditional research focuses on end-to-end communication (unicast) problems [23–25], and its routing algorithms and routing metrics are based on the quality of end-to-end communication. Nowadays, there are also some studies for multicast problems [26–28], but the difference between“unicast” and“multicast” mainly lies in the improvement of routing algorithms but not evaluation metrics. Traditional routing algorithms show some inspirations to the broadcast mechanism, while the evaluation metrics are obviously no longer applicable. This section will propose some frameworks for centralized and distributed routing algorithms under broadcast mechanisms. 3.1. Centralized Routing Algorithm Framework. The fundamental reason why space networks can achieve centralized routing is that they are predictable. And we can directly design the optimal transmission routing table by predicting the future contacting graph. The centralized routing under the broadcast mechanism has the following features: (1) A connection-oriented transmission protocol is usually used, i.e., when two nodes satisfy the connection conditions, they are considered as linked, and all the links are equal. For example, node A establishes a connection with node B, and at the same time node A establishes a connection with node C, the two links are equal in the calculation of the optimal path, regardless of which node is closer to node A. This is because usually the waiting time for nodes to establish a link is much longer than the information propagation time, so the path distance can be ignored 2 Space: Science & Technology (2) Create evolutionary contacting graphs based on model characteristics. Since the source node has to design the optimal path for broadcasting to all other nodes, it needs to maintain a series of evolutionary graphs from current to the future. There are 2 elements need to be considered while designing contacting graphs: (a) The time interval between adjacent graphs (b) The total number of required graphs, which can reflect the information survival time (time-tolive, TTL) These two elements directly affect the storage and computation, and a feasible optimization method is to discard the connection whose duration is shorter than a certain threshold and then select the remaining shortest duration time as the graph interval time. The threshold must be greater than the sum of the message sending delay and propagation delay. The selection method for the number of graphs, i.e., the TTL, is usually related to the network connectivity performance, which will be analyzed in detail later. After above steps, let the distance between nodes be the moment that they connect. The contacting graph of Figure 1 is shown in Figure 2. After obtaining contacting graph as shown in Figure 2, the optimal routing problem can be treated as optimal path problem. (3) Select an optimal path algorithm, usually the shortest path algorithm can be used, such as the Dijkstra algorithm [29]. But our case is a little different from classical problem since there is a time sequency limitation. Fortunately, the Dijkstra algorithm can alwaysfind one shortest path from source node to any others with forward property [30] (i.e., easily satisfying the time sequency requirement with a little modification), and the detail of the improved Dijkstra algorithm for the shortest-path problem in above evolving graphs is discussed in [19] 3.2. Distributed Routing Algorithm Framework. The epidemic algorithm based on theflooding mechanism is an efficient distributed routing algorithm for broadcast [31, 32]. Each time two nodes establish connections, they exchange the list of messages IDs and transmit the messages that the opposite does not have by comparing the list. Obviously, the epidemic model can achieve the fastest message transmission to all nodes, but when the number of nodes is large, connections will be established constantly, thus generating redundant list exchanges and consuming more energy. One solution is to reduce the transmission times by setting a t = t1 t = t2 t = t3 t = t4 C D A B C D B A C D B A C D A B Figure 1: Topology evolution of dynamic network. 3 B 1,3 1,3 2,3 2,3 D C A Figure 2: The contacting graph of Figure 1. Space: Science & Technology 3
第31页
21 framework, routing algorithms, and the effective solutions to unexpected situation [12, 13]. Their main works include designing both the space-time graph and the optimal transmission routing table from the source node to the destination node. Such space-time graph-based algorithms are channel resources friendly, yet they have high computational complexity. Similar space-time graph-based algorithms can be referred to [14–17]. Yan et al. and Wang et al. have evaluated these typical DTN routing algorithms and designed an onboard routing algorithm [18, 19]. Based on the transmission delay, delivery ratio, they have given some suggestions for future space DTN research. Scholars have conducted extensive research on space DTN, and most of these focus on end-to-end information transmission. Future LEO satellite clusters, such like Blackjack program, are more dependent on information sharing among all satellites, so as to achieve efficient data fusion, autonomous decision-making, and collaboration. However, there is still a lack of research on space dynamic networks under this broadcasting mechanism. This paper is aimed at proposing some metrics that can effectively evaluate the network performance and routing algorithm performance under broadcasting mechanism and providing valuable ideas for subsequent research. 2. Model Static network has simpler model, with more complete research in theory and engineering implementations. The dynamic network is totally different since the dynamic network is with sequence and irreversibility. For example, if node A is connecting with node B and B is connecting to C, then there is a link between A and C. However, in dynamic network, if the connection between A and B occurs after B and C, then we cannot assert that there is a link between A and C. Dynamic networks are usually divided into two forms: lossless form and lossy form. The lossless form records all the connection status, while lossy form divides timespan into several segments, regards every segment as a static network, andfinally connects all segments to form a complete network. The lossless form needs to record every moment that connection relationship has changed, which requires a large amount of storage. Therefore, the lossy form can be used to model a dynamic network [12, 13]. Considering a quaternion, c = ð Þ a, b, t,δt , ð1Þ where fa, bg represents the two nodes, respectively, and denotes the moment connection occurring andδt denotes the duration of the connection. For a segment ½tm, tm +Δt�, if c satisfies one of the following conditions, tm≤ t < tm +Δt, tm≤ t +δt < tm +Δt, t≤ tm < tm +Δt < t +δt: ð2Þ Meeting any of the above conditions means that there is always a part of time in the whole connecting duration that is within the specified segment ½tm, tm +Δt�. Then, an edge is created between the nodes, and all such edges form the set of edges Em. Let the snapshot of segment ½tm, tm +Δt� be Sm = ðN, EmÞ. Then, we can model the dynamic network as G = ðI, SÞ, where I represents the total timespan, I = ½Tstart, Tend�, and S is the set of all snapshots. Accumulating each small segment can obtain the entire networks. As shown in Figure 1, there exist four nodes N = fA, B, C, Dg, and the subgraphs of Figure 1 are snapshots fS1, S2, S3, S4g. At time t1, if node A needs to transmit information to target node C, there are two routing methods: (1) information is transmitted from node A to node D at time t1 and then transmitted from node D to node C at t2, and this “store-wait-forward” routing way belongs to the DTN method; (2) node A waits until t3 to establish a complete path with node C, and this routing method is often used for ad hoc networks. Obviously, the complete path between the source and the destination may not always appear, so the DTN routing method can not only realize the information delivery in a shorter time, but also improve the delivery ratio [20–22]. 3. Routing in Space Networks under Broadcasting Mechanism Traditional research focuses on end-to-end communication (unicast) problems [23–25], and its routing algorithms and routing metrics are based on the quality of end-to-end communication. Nowadays, there are also some studies for multicast problems [26–28], but the difference between“unicast” and“multicast” mainly lies in the improvement of routing algorithms but not evaluation metrics. Traditional routing algorithms show some inspirations to the broadcast mechanism, while the evaluation metrics are obviously no longer applicable. This section will propose some frameworks for centralized and distributed routing algorithms under broadcast mechanisms. 3.1. Centralized Routing Algorithm Framework. The fundamental reason why space networks can achieve centralized routing is that they are predictable. And we can directly design the optimal transmission routing table by predicting the future contacting graph. The centralized routing under the broadcast mechanism has the following features: (1) A connection-oriented transmission protocol is usually used, i.e., when two nodes satisfy the connection conditions, they are considered as linked, and all the links are equal. For example, node A establishes a connection with node B, and at the same time node A establishes a connection with node C, the two links are equal in the calculation of the optimal path, regardless of which node is closer to node A. This is because usually the waiting time for nodes to establish a link is much longer than the information propagation time, so the path distance can be ignored 2 Space: Science & Technology (2) Create evolutionary contacting graphs based on model characteristics. Since the source node has to design the optimal path for broadcasting to all other nodes, it needs to maintain a series of evolutionary graphs from current to the future. There are 2 elements need to be considered while designing contacting graphs: (a) The time interval between adjacent graphs (b) The total number of required graphs, which can reflect the information survival time (time-tolive, TTL) These two elements directly affect the storage and computation, and a feasible optimization method is to discard the connection whose duration is shorter than a certain threshold and then select the remaining shortest duration time as the graph interval time. The threshold must be greater than the sum of the message sending delay and propagation delay. The selection method for the number of graphs, i.e., the TTL, is usually related to the network connectivity performance, which will be analyzed in detail later. After above steps, let the distance between nodes be the moment that they connect. The contacting graph of Figure 1 is shown in Figure 2. After obtaining contacting graph as shown in Figure 2, the optimal routing problem can be treated as optimal path problem. (3) Select an optimal path algorithm, usually the shortest path algorithm can be used, such as the Dijkstra algorithm [29]. But our case is a little different from classical problem since there is a time sequency limitation. Fortunately, the Dijkstra algorithm can alwaysfind one shortest path from source node to any others with forward property [30] (i.e., easily satisfying the time sequency requirement with a little modification), and the detail of the improved Dijkstra algorithm for the shortest-path problem in above evolving graphs is discussed in [19] 3.2. Distributed Routing Algorithm Framework. The epidemic algorithm based on theflooding mechanism is an efficient distributed routing algorithm for broadcast [31, 32]. Each time two nodes establish connections, they exchange the list of messages IDs and transmit the messages that the opposite does not have by comparing the list. Obviously, the epidemic model can achieve the fastest message transmission to all nodes, but when the number of nodes is large, connections will be established constantly, thus generating redundant list exchanges and consuming more energy. One solution is to reduce the transmission times by setting a t = t1 t = t2 t = t3 t = t4 C D A B C D B A C D B A C D A B Figure 1: Topology evolution of dynamic network. 3 B 1,3 1,3 2,3 2,3 D C A Figure 2: The contacting graph of Figure 1. Space: Science & Technology 3
第32页
22 reasonable message survival time TTL and deleting expired messages [9]. In addition, employing a priori knowledge of the network can also get better results [33–36], which include the routing algorithms based on the structure of network communities. This algorithm uses different routing mechanism for intercommunity and intracommunity, to achieve better routing results. The use of this methodfirst requires reliable community detection, and for the problem of distributed community detection in dynamic networks, this paper proposes a method that is simple to implement by only three steps: (1) Set all spacecraft as a single-node community and set the similarity between nodes to 0 (2) Calculate the similarity between nodes (3) Compare the similarity with the threshold and then update the community ID according to the comparison The similarity f i,j ðtÞ between node i and node j at time t is updated by the following: f i,j ð Þt = 1 + f i,j ð Þ t− 1⋅ω ð Þ connection exists at t , f i,j ð Þ t− 1⋅ω ð Þ connection not exists at t , ( ð3Þ where the decay coefficientω satisfies 0 <ω < 1. If node i and node j are always connected, the similarity is max f i,j ð Þt h i =1+ω +ω2 +ω3 +⋯: ð4Þ And it satisfies the following: max f i,j ð Þt h i < 1 1−ω : ð5Þ If node i and node j are not connected at moment t, the similarity f i,j ðtÞ is f i,j ð Þt = f i,j ð Þ t− 1⋅ω: ð6Þ And it satisfies the following: f i,j ð Þt < ω 1−ω : ð7Þ Therefore, let the thresholdλ be λ =ω 1−ω : ð8Þ This threshold ensures that members within the same community remain connected when the update time is reached, and update and determine the community ID synchronously. The update strategy is as follows: (1) If f i,j ðtÞ≥λ, then group the nodes i and j into the same community (2) If f i,j ðtÞ <λ, then divide the nodes i and j into different communities (3) When nodes i and j are in the same community and nodes j and k are in the same community, then group the nodes i and k into the same community through simple communication In our experiment, community detection is performed for 115 satellites in orbit, and the community structure at some point is shown as Figure 3. After the distributed community detection, a community structure which is highly connected intracommunity can be obtained. Based on this, a distributed routing framework can be designed as shown in Figure 4. In Figure 4, fA, B, C, D, E, Fg are the IDs of each community, and the specific process of the algorithm is as follows: (1) Intercommunities. Using the epidemic model, the list records the community IDs which have obtained the corresponding message, and when nodes meet, only part of the list is exchanged (opposite is not recorded in this part), and through it to build a delivering request. The opposite accepts (opposite has not received these messages so far) or rejects (opposite has already received these messages but the requester is still unknown, e.g.,“A” sends message to“B,” and then,“A” sends it to“C.” Through the list,“C” knows that“B” has the message yet“B” does not know that“C” already has it.“B” plays the role of “requester,” and“C” is the“opposite” who rejects the requirement) it according to the conditions and updates the list after the delivery is completed. For –6 –4 –2 0 2 4 6 X (m) –6 –4 –2 0 2 4 6 Z (m) × 106 × 106 Figure 3: Community structure of the onboard S/C. 4 Space: Science & Technology example, when any two nodes in community E and community F meet again after the list in Figure 4 updated, no exchange request is made to F because F exists in all lists of community E. Similarly, E exists in all lists of community F, so no exchange request is made to E. (2) Intracommunity. Due to the strong connectivity, it is possible to quickly generate a static graph of the community and design an optimal routing policy without redundancy. In the distributed routing algorithm, nodes do not need to maintain a large amount of contacting graph or design a routing table in advance. It not only saves computational resources, but also has stronger robustness. 4. Performance Evaluation At present, there are few studies on the broadcasting mechanism of space networks with rapidly changing topology. It is necessary to carry out the evaluation mechanism for both network itself and routing algorithm. To this end, we aim to provide theoretical support for the design of future spacebased system and both the routing algorithms. 4.1. Average Delivery Ratio. The average delivery ratio is defined as the average ratio of nodes that receive randomly generated messages within TTL. And it can reflect the connectivity of the entire network. It can be calculated as follows: D =∑s j=1∑n i=1dj i s⋅ ð Þ n− 1 , ð9Þ where s is the total number of messages, n is the number of nodes, and dj i is obtained by the following: dj i = 1 the ð Þ jth message is delivered to node i , 0 node ð Þ i is the source , 0 the ð Þ jth message is not delivered to node i : 8 >>< >>: ð10Þ Ignoring the sending and propagation delay, we can define the instantaneous connectivity of the network as follows: Dinstant = ∑s j=1∑n i=1dj i s⋅ ð Þ n− 1 � � � � � TTL=0 : ð11Þ 4.2. Average Transmission Time. The average transmission time can reflect the degree of network connectivity and is defined as the average time that expended for delivering messages to other nodes within the message survival time TTL. To describe the performance more concretely, we use relative average transmission time and absolute average transmission time, respectively. 4.2.1. Relative Average Transmission Time. The relative average transmission time can be defined as follows: Trela =∑s j=1∑n i=1t j i s⋅ ð Þ n− 1 , ð12Þ Msg 1 Msg 1 2 List While: A E Msg 1 List C F B Msg 1 2 List A C F B F Msg 1 2 List A C E B E Intra– community routing Intra– community routing Msg 2 Msg 1 E A B C F Connecting Figure 4: Routing based on community structure. Space: Science & Technology 5
第33页
23 reasonable message survival time TTL and deleting expired messages [9]. In addition, employing a priori knowledge of the network can also get better results [33–36], which include the routing algorithms based on the structure of network communities. This algorithm uses different routing mechanism for intercommunity and intracommunity, to achieve better routing results. The use of this methodfirst requires reliable community detection, and for the problem of distributed community detection in dynamic networks, this paper proposes a method that is simple to implement by only three steps: (1) Set all spacecraft as a single-node community and set the similarity between nodes to 0 (2) Calculate the similarity between nodes (3) Compare the similarity with the threshold and then update the community ID according to the comparison The similarity f i,j ðtÞ between node i and node j at time t is updated by the following: f i,j ð Þt = 1 + f i,j ð Þ t− 1⋅ω ð Þ connection exists at t , f i,j ð Þ t− 1⋅ω ð Þ connection not exists at t , ( ð3Þ where the decay coefficientω satisfies 0 <ω < 1. If node i and node j are always connected, the similarity is max f i,j ð Þt h i =1+ω +ω2 +ω3 +⋯: ð4Þ And it satisfies the following: max f i,j ð Þt h i < 1 1−ω : ð5Þ If node i and node j are not connected at moment t, the similarity f i,j ðtÞ is f i,j ð Þt = f i,j ð Þ t− 1⋅ω: ð6Þ And it satisfies the following: f i,j ð Þt < ω 1−ω : ð7Þ Therefore, let the thresholdλ be λ =ω 1−ω : ð8Þ This threshold ensures that members within the same community remain connected when the update time is reached, and update and determine the community ID synchronously. The update strategy is as follows: (1) If f i,j ðtÞ≥λ, then group the nodes i and j into the same community (2) If f i,j ðtÞ <λ, then divide the nodes i and j into different communities (3) When nodes i and j are in the same community and nodes j and k are in the same community, then group the nodes i and k into the same community through simple communication In our experiment, community detection is performed for 115 satellites in orbit, and the community structure at some point is shown as Figure 3. After the distributed community detection, a community structure which is highly connected intracommunity can be obtained. Based on this, a distributed routing framework can be designed as shown in Figure 4. In Figure 4, fA, B, C, D, E, Fg are the IDs of each community, and the specific process of the algorithm is as follows: (1) Intercommunities. Using the epidemic model, the list records the community IDs which have obtained the corresponding message, and when nodes meet, only part of the list is exchanged (opposite is not recorded in this part), and through it to build a delivering request. The opposite accepts (opposite has not received these messages so far) or rejects (opposite has already received these messages but the requester is still unknown, e.g.,“A” sends message to“B,” and then,“A” sends it to“C.” Through the list,“C” knows that“B” has the message yet“B” does not know that“C” already has it.“B” plays the role of “requester,” and“C” is the“opposite” who rejects the requirement) it according to the conditions and updates the list after the delivery is completed. For –6 –4 –2 0 2 4 6 X (m) –6 –4 –2 0 2 4 6 Z (m) × 106 × 106 Figure 3: Community structure of the onboard S/C. 4 Space: Science & Technology example, when any two nodes in community E and community F meet again after the list in Figure 4 updated, no exchange request is made to F because F exists in all lists of community E. Similarly, E exists in all lists of community F, so no exchange request is made to E. (2) Intracommunity. Due to the strong connectivity, it is possible to quickly generate a static graph of the community and design an optimal routing policy without redundancy. In the distributed routing algorithm, nodes do not need to maintain a large amount of contacting graph or design a routing table in advance. It not only saves computational resources, but also has stronger robustness. 4. Performance Evaluation At present, there are few studies on the broadcasting mechanism of space networks with rapidly changing topology. It is necessary to carry out the evaluation mechanism for both network itself and routing algorithm. To this end, we aim to provide theoretical support for the design of future spacebased system and both the routing algorithms. 4.1. Average Delivery Ratio. The average delivery ratio is defined as the average ratio of nodes that receive randomly generated messages within TTL. And it can reflect the connectivity of the entire network. It can be calculated as follows: D =∑s j=1∑n i=1dj i s⋅ ð Þ n− 1 , ð9Þ where s is the total number of messages, n is the number of nodes, and dj i is obtained by the following: dj i = 1 the ð Þ jth message is delivered to node i , 0 node ð Þ i is the source , 0 the ð Þ jth message is not delivered to node i : 8 >>< >>: ð10Þ Ignoring the sending and propagation delay, we can define the instantaneous connectivity of the network as follows: Dinstant = ∑s j=1∑n i=1dj i s⋅ ð Þ n− 1 � � � � � TTL=0 : ð11Þ 4.2. Average Transmission Time. The average transmission time can reflect the degree of network connectivity and is defined as the average time that expended for delivering messages to other nodes within the message survival time TTL. To describe the performance more concretely, we use relative average transmission time and absolute average transmission time, respectively. 4.2.1. Relative Average Transmission Time. The relative average transmission time can be defined as follows: Trela =∑s j=1∑n i=1t j i s⋅ ð Þ n− 1 , ð12Þ Msg 1 Msg 1 2 List While: A E Msg 1 List C F B Msg 1 2 List A C F B F Msg 1 2 List A C E B E Intra– community routing Intra– community routing Msg 2 Msg 1 E A B C F Connecting Figure 4: Routing based on community structure. Space: Science & Technology 5
第34页
24 where t j i is obtained by the following: t j i = TDj i ð Þ the jth message is delivered to node i , TTL the ð Þ jth message is not delivered to node i , ( ð13Þ where TDj i is the time delay of jth message delivered to node i. 4.2.2. Absolute Average Transmission Time. The absolute average transmission time can be defined as follows: Tabs =∑s j=1∑n i=1t j i ∑s j=1 nj− 1� � , ð14Þ where t j i is obtained by the following: t j i = TDj i ð Þ the jth message is delivered to node i , 0 the ð Þ jth message is not delivered to node i : ( ð15Þ 4.3. Average Times of RREQ Initiated. To evaluate the efficiency of the routing algorithm, the metric of the average times of RREQ initiated is proposed. Since the routing goal of the space network is to transmit messages to all nodes, then the total amount of messages that need to be transmitted is equal for both theflooding algorithm and the optimal routing algorithm. The difference between them lies in the number of connection establishment and the times of routing requires initiated. Theflooding algorithm needs to initiate aflood of routing requests, while the optimal routing algorithm only needs to initiate n− 1 times routing requests. Obviously, it is possible to reduce energy consumption by designing efficient routing algorithms that reduce the times of RREQ initiated. We define the average times of RREQ initiated as follows: H =∑n−1 i=1∑n j=i+1hj i s⋅ ð Þ n− 1⋅ D , ð16Þ where s is the total number of messages, n is the number of nodes, D is the average delivery ratio, and hj i is the times of RREQ initiated between node i and node j. 4.4. Ratio of Congested Node. The ratio of congested nodes can reflect the engineering feasibility of the routing algorithm. If the node transmission capacity is not considered, the designed routing algorithm may excessively use a specific node in a short period of time, which leads to node overload as well as data loss. Therefore, designing the routing algorithm by considering the congestion of nodes will enhance the reliability of the algorithm in practical engineering. We define it as the ratio of nodes that arise congestion events when messages are broadcast to the entire network: C =∑n i=1ci n , ð17Þ where ci is defined as follows: ci = 1 ð Þ U≥λ , 0 ð Þ U <λ , ( ð18Þ and where U is the ratio of the messages need to be transmitted by node i to the maximum transmission capacity of node i andλ is the capability threshold, where U is related to both the total number of messages, the message generation interval, and the routing algorithm. 5. Simulation and Analysis Combined with the orbit of space-based systems, we implement simulation to verify the rationality of the metrics that we proposed for evaluating the performance of dynamic network and the broadcasting mechanism. The simulation uses the actual orbital data of infrared and electrodetection satellites in orbit. The communication conditions are set as Figure 5. Two satellites can realize communicate when they meet the following conditions (approximately estimated according to Iridium constellation [37, 38]): (a) the distance between them is less than 4000 km; (b) the height of communication link is always larger than 100 km to the ground. 5.1. Performance Comparison of Different Space Networks. We have selected 10 satellites randomly among all satellites to generate a satellite network, and the network is expanded based on add new satellites. Allfive groups of the networks are shown in Figure 6. The simulation generates 500 messages randomly in multiple moments, respectively, and sets the TTL to 3000 seconds. To compare the performance of different network Satellite 1 d < 4000 km Satellite 2 h > 100 km 0 Figure 5: Schematic of intersatellite communicable conditions. 6 Space: Science & Technology structure, we both use epidemic routing algorithm for these networks. Figure 7 shows the average delivery ratio of networks in Figure 6. The connectivity performance of the network tends to increase as the number of satellites in the network increases and the average delivery ratio keeps increasing at the same TTL. The absolute average transmission time for the networks in Figure 6 is shown in Figure 8. The relative average transmission time for the networks in Figure 6 is shown in Figure 9. The calculation of absolute transmission time is independent from the delivery ratio and only considers the time spent on the actual delivered messages, so it needs to be used together with the average delivery ratio when measuring the network performance. The calculation of relative transmission time considers the undelivered messages and configures their time as TTL, so the relative transmission time also reflects the delivery ratio to a certain extent. The average delivery ratio of the network reflects the connectivity of the network, and the average transmission time reflects the degree of connectivity, so that the performance of the network can be reliably evaluated using these two metrics. 5.2. Centralized vs. Distributed Routing Algorithm. The simulation randomly selects a 50,000-second timespan and generates messages at random nodes and random moments. Then, we use the centralized and distributed routing algorithms proposed previously for message routing. In the centralized algorithm, the interval of adjacent contacting graphs is 300 seconds, and 10 future graphs are maintained by each node. The delivery ratio of thefive group networks in Figure 6 under the two algorithms is shown in Figure 10. The delivery ratio of the distributed algorithm is almost equal to the centralized algorithm. The relative average transmission time of thefive groups of networks in Figure 6 under the two algorithms is shown in Figure 11. The centralized algorithm has a slightly shorter transmission time than the distributed algorithm. The average times of RREQ initiated for thefive groups of networks in Figure 6 under the two algorithms are shown in Figure 12. And the total number of messages is equal. Under the centralized algorithm, each message is transmitted according to its own routing table, so the average times of RREQ initiated are always 1, independent of whether it is successfully delivered or not and both independent of the density of message. Under the distributed algorithm, when the message is dense, the transmission confirmation of multiple messages can be completed simultaneously in one route Net-1 S/C number: 10 Net-2 S/C number: 19 Net-4 S/C number: 37 Net-5 S/C number: 43 Net-3 S/C number: 29 Figure 6: Five groups of satellite networks. 2 4 6 8 10 Time (s) × 104 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Avg. delivery ratio Net-1 Net-2 Net-3 Net-4 Net-5 Figure 7: Average delivery ratio of networks in Figure 6. Space: Science & Technology 7
第35页
25 where t j i is obtained by the following: t j i = TDj i ð Þ the jth message is delivered to node i , TTL the ð Þ jth message is not delivered to node i , ( ð13Þ where TDj i is the time delay of jth message delivered to node i. 4.2.2. Absolute Average Transmission Time. The absolute average transmission time can be defined as follows: Tabs =∑s j=1∑n i=1t j i ∑s j=1 nj− 1� � , ð14Þ where t j i is obtained by the following: t j i = TDj i ð Þ the jth message is delivered to node i , 0 the ð Þ jth message is not delivered to node i : ( ð15Þ 4.3. Average Times of RREQ Initiated. To evaluate the efficiency of the routing algorithm, the metric of the average times of RREQ initiated is proposed. Since the routing goal of the space network is to transmit messages to all nodes, then the total amount of messages that need to be transmitted is equal for both theflooding algorithm and the optimal routing algorithm. The difference between them lies in the number of connection establishment and the times of routing requires initiated. Theflooding algorithm needs to initiate aflood of routing requests, while the optimal routing algorithm only needs to initiate n− 1 times routing requests. Obviously, it is possible to reduce energy consumption by designing efficient routing algorithms that reduce the times of RREQ initiated. We define the average times of RREQ initiated as follows: H =∑n−1 i=1∑n j=i+1hj i s⋅ ð Þ n− 1⋅ D , ð16Þ where s is the total number of messages, n is the number of nodes, D is the average delivery ratio, and hj i is the times of RREQ initiated between node i and node j. 4.4. Ratio of Congested Node. The ratio of congested nodes can reflect the engineering feasibility of the routing algorithm. If the node transmission capacity is not considered, the designed routing algorithm may excessively use a specific node in a short period of time, which leads to node overload as well as data loss. Therefore, designing the routing algorithm by considering the congestion of nodes will enhance the reliability of the algorithm in practical engineering. We define it as the ratio of nodes that arise congestion events when messages are broadcast to the entire network: C =∑n i=1ci n , ð17Þ where ci is defined as follows: ci = 1 ð Þ U≥λ , 0 ð Þ U <λ , ( ð18Þ and where U is the ratio of the messages need to be transmitted by node i to the maximum transmission capacity of node i andλ is the capability threshold, where U is related to both the total number of messages, the message generation interval, and the routing algorithm. 5. Simulation and Analysis Combined with the orbit of space-based systems, we implement simulation to verify the rationality of the metrics that we proposed for evaluating the performance of dynamic network and the broadcasting mechanism. The simulation uses the actual orbital data of infrared and electrodetection satellites in orbit. The communication conditions are set as Figure 5. Two satellites can realize communicate when they meet the following conditions (approximately estimated according to Iridium constellation [37, 38]): (a) the distance between them is less than 4000 km; (b) the height of communication link is always larger than 100 km to the ground. 5.1. Performance Comparison of Different Space Networks. We have selected 10 satellites randomly among all satellites to generate a satellite network, and the network is expanded based on add new satellites. Allfive groups of the networks are shown in Figure 6. The simulation generates 500 messages randomly in multiple moments, respectively, and sets the TTL to 3000 seconds. To compare the performance of different network Satellite 1 d < 4000 km Satellite 2 h > 100 km 0 Figure 5: Schematic of intersatellite communicable conditions. 6 Space: Science & Technology structure, we both use epidemic routing algorithm for these networks. Figure 7 shows the average delivery ratio of networks in Figure 6. The connectivity performance of the network tends to increase as the number of satellites in the network increases and the average delivery ratio keeps increasing at the same TTL. The absolute average transmission time for the networks in Figure 6 is shown in Figure 8. The relative average transmission time for the networks in Figure 6 is shown in Figure 9. The calculation of absolute transmission time is independent from the delivery ratio and only considers the time spent on the actual delivered messages, so it needs to be used together with the average delivery ratio when measuring the network performance. The calculation of relative transmission time considers the undelivered messages and configures their time as TTL, so the relative transmission time also reflects the delivery ratio to a certain extent. The average delivery ratio of the network reflects the connectivity of the network, and the average transmission time reflects the degree of connectivity, so that the performance of the network can be reliably evaluated using these two metrics. 5.2. Centralized vs. Distributed Routing Algorithm. The simulation randomly selects a 50,000-second timespan and generates messages at random nodes and random moments. Then, we use the centralized and distributed routing algorithms proposed previously for message routing. In the centralized algorithm, the interval of adjacent contacting graphs is 300 seconds, and 10 future graphs are maintained by each node. The delivery ratio of thefive group networks in Figure 6 under the two algorithms is shown in Figure 10. The delivery ratio of the distributed algorithm is almost equal to the centralized algorithm. The relative average transmission time of thefive groups of networks in Figure 6 under the two algorithms is shown in Figure 11. The centralized algorithm has a slightly shorter transmission time than the distributed algorithm. The average times of RREQ initiated for thefive groups of networks in Figure 6 under the two algorithms are shown in Figure 12. And the total number of messages is equal. Under the centralized algorithm, each message is transmitted according to its own routing table, so the average times of RREQ initiated are always 1, independent of whether it is successfully delivered or not and both independent of the density of message. Under the distributed algorithm, when the message is dense, the transmission confirmation of multiple messages can be completed simultaneously in one route Net-1 S/C number: 10 Net-2 S/C number: 19 Net-4 S/C number: 37 Net-5 S/C number: 43 Net-3 S/C number: 29 Figure 6: Five groups of satellite networks. 2 4 6 8 10 Time (s) × 104 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Avg. delivery ratio Net-1 Net-2 Net-3 Net-4 Net-5 Figure 7: Average delivery ratio of networks in Figure 6. Space: Science & Technology 7
第36页
26 request; when the message is sparse, the transmission confirmation of multiple messages cannot be completed simultaneously. Due to theflooding mechanism for messages 0 500 1000 1500 2000 2500 3000 3500 4000 Messages generation interval (s) 0 0.5 1 1.5 2 2.5 3 3.5 4 Avg. RREQ times Net-1-distributed Net-2-distributed Net-3-distributed Net-4-distributed Net-5-distributed Net-1 to 5-centralized Figure 12: Average times of RREQ initiated under centralized and distributed algorithms. 0 1000 2000 3000 4000 5000 6000 7000 8000 TTL (s) 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Avg. delivery ratio Net-1 Net-2 Net-3 Net-4 Net-5 Net-6 Figure 13: Average delivery ratio of multigroup network with different TTL. Net-1 Avg. delivery ratio Net-2 Net-3 Net-4 Net-5 Centralized Distributed 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Figure 10: Delivery ratio under centralized and distributed algorithms. Net-1 Rel. avg. trans. time (s) Net-2 Net-3 Net-4 Net-5 Centralized Distributed 2500 2000 1500 1000 500 0 Figure 11: Relative transmission time under centralized and distributed algorithms. 2 4 6 8 10 Time (s) × 104 0 1600 Abs. avg. trans. time (s) Net-1 Net-2 Net-3 Net-4 Net-5 200 400 600 800 1000 1200 1400 Figure 8: Absolute average transmission time of networks in Figure 6. 2 4 6 8 10 Time (s) × 104 0 2500 Rel. avg. trans. time (s) Net-1 Net-2 Net-3 Net-4 Net-5 500 1000 1500 2000 Figure 9: Relative average transmission time of networks in Figure 6. 8 Space: Science & Technology delivering intercommunity, and the route request cannot be completed for multiple messages at the same time, the average times of RREQ initiated gradually rise when the messages are sparse. And the higher the number of nodes in the network, the larger the average times of RREQ initiated will be. 5.3. Message TTL Setting. The message survival time TTL directly affects the message delivery ratio. Setting the TTL too small will lead to low delivery ratio, and too large will lead to communication resource wasting since some routing algorithms (e.g., epidemic algorithm) will keep initiating routing requests. Therefore, this subsection conducts a simulation on the TTL and proposes some suggestions. Six groups of networks with different connectivity performance are generated and simulated, and messages are generated at random nodes and random moments. We use the epidemic algorithm for message routing, and the variation of delivery ratio with different TTL is shown in Figure 13. As the TTL increases, the delivery ratio gradually increases to 1. And when the TTL is 0, it returns the instantaneous connectivity Dinstant defined in the previous section, and the larger the Dinstant, the smaller the TTL required for a demanding delivery ratio. This section is aimed at calculating the specific TTL value by only. To this end, it is necessary tofit the delivery ratio curve with respect to both and TTL. The ratio curve isfitted as follows: y = 1 1 + 1/ ð Þ Dinstant− 1⋅ exp−TTL⋅ Dinstant ð Þ /350 , ð19Þ where y denotes the delivery ratio. Figure 14 shows thefitted and real curves. By obtaining the instantaneous connectivity in advance through ground simulation and setting the required delivery ratio to 0.99, the required TTL can be calculated by follows: TTL = 350 Dinstant ⋅ ln 1/Dinstant− 1 1/0:99− 1   : ð20Þ The star at Figure 14 is the TTL time required for 99% delivery ratio calculated according to the above formula. 6. Conclusions In this paper, the space network under the broadcasting mechanism is studied. Firstly, we have given definition and analysis of the dynamic network model. Then, the space routing algorithm framework under broadcast mechanism is introduced, which includes centralized routing algorithm based on the model predictability and distributed algorithm based on community structure. Moreover, the performance evaluation indicators of network and routing algorithm are proposed, including average delivery ratio, average transmission time, average times of RREQ, and the ratio of congested nodes. Simulation is conducted for several groups of space networks, respectively, and the evaluation indicators can effectively distinguish the good and bad network structure. Results show the reasonableness of the evaluation metrics. 20000 0.5 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 20000 0.4 0.5 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 0.4 0.6 Avg. delivery ratio 0.8 8000 20000 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 20000 0.2 0.4 0.6 0.8 1 4000 6000 Real Fitted 10000 20000 0.2 1 0.4 0.6 0.8 0.2 1 4000 6000 TTL (s) 8000 Real Fitted 10000 20000 4000 6000 8000 Real Fitted 10000 Figure 14: Fitted curves vs. real curves. Space: Science & Technology 9
第37页
27 request; when the message is sparse, the transmission confirmation of multiple messages cannot be completed simultaneously. Due to theflooding mechanism for messages 0 500 1000 1500 2000 2500 3000 3500 4000 Messages generation interval (s) 0 0.5 1 1.5 2 2.5 3 3.5 4 Avg. RREQ times Net-1-distributed Net-2-distributed Net-3-distributed Net-4-distributed Net-5-distributed Net-1 to 5-centralized Figure 12: Average times of RREQ initiated under centralized and distributed algorithms. 0 1000 2000 3000 4000 5000 6000 7000 8000 TTL (s) 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Avg. delivery ratio Net-1 Net-2 Net-3 Net-4 Net-5 Net-6 Figure 13: Average delivery ratio of multigroup network with different TTL. Net-1 Avg. delivery ratio Net-2 Net-3 Net-4 Net-5 Centralized Distributed 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Figure 10: Delivery ratio under centralized and distributed algorithms. Net-1 Rel. avg. trans. time (s) Net-2 Net-3 Net-4 Net-5 Centralized Distributed 2500 2000 1500 1000 500 0 Figure 11: Relative transmission time under centralized and distributed algorithms. 2 4 6 8 10 Time (s) × 104 0 1600 Abs. avg. trans. time (s) Net-1 Net-2 Net-3 Net-4 Net-5 200 400 600 800 1000 1200 1400 Figure 8: Absolute average transmission time of networks in Figure 6. 2 4 6 8 10 Time (s) × 104 0 2500 Rel. avg. trans. time (s) Net-1 Net-2 Net-3 Net-4 Net-5 500 1000 1500 2000 Figure 9: Relative average transmission time of networks in Figure 6. 8 Space: Science & Technology delivering intercommunity, and the route request cannot be completed for multiple messages at the same time, the average times of RREQ initiated gradually rise when the messages are sparse. And the higher the number of nodes in the network, the larger the average times of RREQ initiated will be. 5.3. Message TTL Setting. The message survival time TTL directly affects the message delivery ratio. Setting the TTL too small will lead to low delivery ratio, and too large will lead to communication resource wasting since some routing algorithms (e.g., epidemic algorithm) will keep initiating routing requests. Therefore, this subsection conducts a simulation on the TTL and proposes some suggestions. Six groups of networks with different connectivity performance are generated and simulated, and messages are generated at random nodes and random moments. We use the epidemic algorithm for message routing, and the variation of delivery ratio with different TTL is shown in Figure 13. As the TTL increases, the delivery ratio gradually increases to 1. And when the TTL is 0, it returns the instantaneous connectivity Dinstant defined in the previous section, and the larger the Dinstant, the smaller the TTL required for a demanding delivery ratio. This section is aimed at calculating the specific TTL value by only. To this end, it is necessary tofit the delivery ratio curve with respect to both and TTL. The ratio curve isfitted as follows: y = 1 1 + 1/ ð Þ Dinstant− 1⋅ exp−TTL⋅ Dinstant ð Þ /350 , ð19Þ where y denotes the delivery ratio. Figure 14 shows thefitted and real curves. By obtaining the instantaneous connectivity in advance through ground simulation and setting the required delivery ratio to 0.99, the required TTL can be calculated by follows: TTL = 350 Dinstant ⋅ ln 1/Dinstant− 1 1/0:99− 1  : ð20Þ The star at Figure 14 is the TTL time required for 99% delivery ratio calculated according to the above formula. 6. Conclusions In this paper, the space network under the broadcasting mechanism is studied. Firstly, we have given definition and analysis of the dynamic network model. Then, the space routing algorithm framework under broadcast mechanism is introduced, which includes centralized routing algorithm based on the model predictability and distributed algorithm based on community structure. Moreover, the performance evaluation indicators of network and routing algorithm are proposed, including average delivery ratio, average transmission time, average times of RREQ, and the ratio of congested nodes. Simulation is conducted for several groups of space networks, respectively, and the evaluation indicators can effectively distinguish the good and bad network structure. Results show the reasonableness of the evaluation metrics. 20000 0.5 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 20000 0.4 0.5 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 0.4 0.6 Avg. delivery ratio 0.8 8000 20000 0.6 0.7 0.8 0.9 1 4000 6000 8000 Real Fitted 10000 20000 0.2 0.4 0.6 0.8 1 4000 6000 Real Fitted 10000 20000 0.2 1 0.4 0.6 0.8 0.2 1 4000 6000 TTL (s) 8000 Real Fitted 10000 20000 4000 6000 8000 Real Fitted 10000 Figure 14: Fitted curves vs. real curves. Space: Science & Technology 9
第38页
28 And then, we implement simulation and analysis of centralized and distributed routing algorithms. Finally, the setting of message survival time TTL, which affects the message delivery ratio, is studied, and the setting suggestion of TTL is given. The simulation of congestion is not given in this paper because the definition of congestion is related to the actual capability of node and the actual data size. Furthermore, the parameters of such indicator need to be set for specific problems; thus, it is not analyzed experimentally in this paper. However, the node congestion is a case that must be considered in practical engineering, so that the routing algorithm can be optimized by the congestion status. This paper presents some problems and feasible research directions in relatedfields. It may help to build the framework for the research of information sharing in dynamic space-based networks. Data Availability The data used to support thefindings of this study are available from the author upon reasonable request. Conflicts of Interest The authors declare that they have no conflicts of interest. Authors’ Contributions Qingrui Zhou contributed to the conception of the study and performed the analysis with constructive discussions. Zipeng Ye performed the experiment and analyzed the results . The writing of the manuscript was done by Zhou and Ye together. Acknowledgments This research was supported by the National Key R&D Program of China under Grant 2018YFA0703800. References [1] J. Foust,“SpaceX’s space-internet woes: despite technical glitches, the company plans to launch thefirst of nearly 12,000 satellites in 2019,” IEEE Spectrum, vol. 56, no. 1, pp. 50-51, 2019. [2] C. R. Boshuizen, J. Mason, P. Klupar, and S. Spanhake, “Results from the planet labsflock constellation,” in 28th Annual AIAA/USU Conference on Small Satellites, Logan, Utah, USA, 2014. [3] J. Q. Zhai and X. F. Li,“Introduction of OneWeb system and domestic LEO internet satellite system,” Space Electronic Technology, vol. 14, no. 6, pp. 1–7, 2017. [4] S. Burleigh, A. Hooke, L. Torgerson et al.,“Delay-tolerant networking: an approach to interplanetary internet,” IEEE Communications Magazine, vol. 41, no. 6, pp. 128–136, 2003. [5] K. Fall,“A delay-tolerant network architecture for challenged internets,” Computer Communication Review, vol. 33, no. 4, pp. 27–34, 2003. [6] A. Vahdat and D. Becker, Epidemic routing for partially connected ad hoc networks, Technical Report CS-200006, Duke University, 2000. [7] A. Lindgren, A. Doria, and O. Schelen,“Probabilistic routing in intermittently connected networks,” ACM SIGMOBILE Mobile Computing Communication Review, vol. 7, no. 3, pp. 19-20, 2003. [8] J. A. Davis, A. H. Fagg, and B. N. Levine,“Wearable computers as packet transport mechanisms in highly-partitioned ad-hoc networks,” in Proceedings Fifth International Symposium on Wearable Computers, Zurich, Switzerland, 2001. [9] K. A. Harras, K. C. Almeroth, and E. M. Belding-Royer,“Delay tolerant mobile networks (DTMNS): controlledflooding in sparse mobile networks,” in NETWORKING 2005. Networking Technologies, Services, and Protocols; Performance of Computer and Communication Networks; Mobile and Wireless Communications Systems. NETWORKING 2005, R. Boutaba, K. Almeroth, R. Puigjaner, S. Shen, and J. P. Black, Eds., vol. 3462 of Lecture Notes in Computer Science, Springer, Berlin, Heidelberg, 2004. [10] H. Dubois-Ferriere, M. Grossglauser, and M. Vetterli,“Age matters: efficient route discovery in mobile ad hoc networks using encounter ages,” in Proceedings of the 4th ACM international symposium on Mobile ad hoc networking & computing - MobiHoc '03, Annapolis, Maryland, USA, 2003. [11] K. Tan, Z. Qian, and W. Zhu,“Shortest path routing in partially connected ad hoc networks,” in GLOBECOM '03. IEEE Global Telecommunications Conference (IEEE Cat. No.03CH37489), San Francisco, CA, USA, 2003. [12] S. Merugu and E. W. Zegura, Routing in space and time in networks with predictable mobility, Technical Report, GIT-CC04-07, Georgia Tech College of Computing, 2004. [13] D. Fischer, D. Basin, and T. Engel,“Topology dynamics and routing for predictable mobile networks,” in 2008 IEEE International Conference on Network Protocols, Orlando, FL, USA, 2008. [14] S. Iranmanesh and K. W. Chin,“A novel mobility-based routing protocol for semi-predictable disruption tolerant networks,” International Journal of Wireless Information Networks, vol. 22, no. 2, pp. 138–146, 2015. [15] C. Liu and J. Wu,“Routing in a cyclic MobiSpace,” in Proceedings of the 9th ACM international symposium on Mobile ad hoc networking and computing - MobiHoc '08, Hong Kong SAR, China, 2008. [16] M. Huang, S. Chen, Z. Ying, and Y. Wang,“Cost-efficient topology design problem in time-evolving delay-tolerant networks,” in 2010 IEEE Global Telecommunications Conference GLOBECOM 2010, Miami, FL, USA, 2010. [17] M. Huang, S. Chen, Y. Zhu, and Y. Wang,“Topology control for time-evolving and predictable delay-tolerant networks,” IEEE Transactions on Computers, vol. 62, no. 11, pp. 2308–2321, 2013. [18] H. C. Yan, J. Guo, and H. J. Zhang,“Performance evaluation of routing algorithms on space delay/disruption tolerant networks,” Chinese Space Science and Technology, vol. 36, no. 4, pp. 38–46, 2016. [19] Y. Wang, B. Liu, W. Yu, and B. Zhao,“Routing algorithm for navigation constellation based on evolving graph model,” Chinese Space Science and Technology, vol. 32, no. 5, pp. 76–83, 2012. [20] S. Jain, K. Fall, and R. Patra,“Routing in a delay tolerant network,” in Proceedings of the ACM SIGCOMM, Portland, Oregon, USA, 2004. 10 Space: Science & Technology [21] J. Ott, D. Kutscher, and C. Dwertmann,“Integrating DTN and MANET routing,”in Proceedings of the 2006 SIGCOMM workshop on Challenged networks - CHANTS '06, Pisa, Italy, 2006. [22] D. Yu and Y. Ko,“FFRDV: fastest-ferry routing in DTN-enabled vehicular ad hoc networks,” in 2009 11th International Conference on Advanced Communication Technology, Gangwon, Korea (South), 2009. [23] E. P. C. Jones, L. Li, J. K. Schmidtke, and P. Ward,“Practical routing in delay-tolerant networks,” IEEE Transaction on Mobile Computing, vol. 6, no. 8, pp. 943–959, 2007. [24] J. Leguay, T. Friedman, and V. Conan,“Evaluating mobility pattern space routing for DTNs,” in Proceedings IEEE INFOCOM 2006. 25TH IEEE International Conference on Computer Communications, Barcelona, Spain, 2005. [25] S. Iranmanesh, R. Raad, and K. W. Chin,“A novel destination-based routing protocol (DBRP) in DTNs,” in 2012 International Symposium on Communications and Information Technologies (ISCIT), Gold Coast, QLD, Australia, 2012. [26] J. Wu and Y. Wang,“A non-replication multicasting scheme in delay tolerant networks,” in The 7th IEEE International Conference on Mobile Ad-hoc and Sensor Systems (IEEE MASS 2010), San Francisco, CA, USA, 2010. [27] L. Junhai, Y. Danxia, X. Liu, and F. Mingyu,“A survey of multicast routing protocols for mobile ad-hoc networks,” IEEE Communications Surveys & Tutorials, vol. 11, no. 1, pp. 78– 91, 2009. [28] L. Junhai, X. Liu, and Y. Danxia,“Research on multicast routing protocols for mobile ad-hoc networks,” Computer Networks, vol. 52, no. 5, pp. 988–997, 2008. [29] E. W. Dijkstra,“A note on two problems in connexion with graphs,” Numerische Mathematik, vol. 1, no. 1, pp. 269–271, 1959. [30] B. B. Xuan, A. Ferreira, and A. Jarry,“Evolving graphs and least cost journeys in dynamic networks,” in WiOpt'03: Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, INRIA Sophia-Antipolis, France, 2003. [31] S. Eshghi, M. H. R. Khouzani, S. Sarkar, N. B. Shroff, and S. S. Venkatesh,“Optimal energy-aware epidemic routing in DTNs,” IEEE Transactions on Automatic Control, vol. 60, no. 6, pp. 1554–1569, 2015. [32] P. Mundur, M. Seligman, and G. Lee,“Epidemic routing with immunity in delay tolerant networks,” in MILCOM 2008 - 2008 IEEE Military Communications Conference, San Diego, CA, USA, 2008. [33] E. Bulut and B. K. Szymanski,“Friendship based routing in delay tolerant mobile social networks,” in 2010 IEEE Global Telecommunications Conference GLOBECOM 2010, Miami, FL, USA, 2010. [34] E. Bulut and B. K. Szymanski,“Exploiting friendship relations for efficient routing in mobile social networksExploiting Friendship Relations for Efficient Routing in Mobile Social Networks,” IEEE Transaction on Parallel and Distributed System, vol. 23, no. 12, pp. 2254–2265, 2012. [35] K. Chen and H. Shen,“SMART: lightweight distributed social map based routing in delay tolerant networks,” , Austin, TX, USA, 2012 20th IEEE International Conference on Network Protocols (ICNP), 2012. [36] P. Hui, E. Yoneki, S. Y. Chan, and J. Crowcroft,“Distributed community detection in delay tolerant networks,” in MobiArch '07: Proceedings of 2nd ACM/IEEE international workshop on Mobility in the evolving internet architecture, Kyoto, Japan, 2007. [37] D. E. Sterling and J. E. Hatlelid,“The IRIDIUM system-a revolutionary satellite communications system developed with innovative applications of technology,” in MILCOM 91 - Conference record, McLean, VA, USA, 1991. [38] C. E. Fossa, R. A. Raines, G. H. Gunsch, and M. A. Temple, “An overview of the IRIDIUM (R) low Earth orbit (LEO) satellite system,” in Proceedings of the IEEE 1998 National Aerospace and Electronics Conference. NAECON 1998. Celebrating 50 Years (Cat. No.98CH36185), Dayton, OH, USA, 1998. Space: Science & Technology 11
第39页
29 And then, we implement simulation and analysis of centralized and distributed routing algorithms. Finally, the setting of message survival time TTL, which affects the message delivery ratio, is studied, and the setting suggestion of TTL is given. The simulation of congestion is not given in this paper because the definition of congestion is related to the actual capability of node and the actual data size. Furthermore, the parameters of such indicator need to be set for specific problems; thus, it is not analyzed experimentally in this paper. However, the node congestion is a case that must be considered in practical engineering, so that the routing algorithm can be optimized by the congestion status. This paper presents some problems and feasible research directions in relatedfields. It may help to build the framework for the research of information sharing in dynamic space-based networks. Data Availability The data used to support thefindings of this study are available from the author upon reasonable request. Conflicts of Interest The authors declare that they have no conflicts of interest. Authors’ Contributions Qingrui Zhou contributed to the conception of the study and performed the analysis with constructive discussions. Zipeng Ye performed the experiment and analyzed the results . The writing of the manuscript was done by Zhou and Ye together. Acknowledgments This research was supported by the National Key R&D Program of China under Grant 2018YFA0703800. References [1] J. Foust,“SpaceX’s space-internet woes: despite technical glitches, the company plans to launch thefirst of nearly 12,000 satellites in 2019,” IEEE Spectrum, vol. 56, no. 1, pp. 50-51, 2019. [2] C. R. Boshuizen, J. Mason, P. Klupar, and S. Spanhake, “Results from the planet labsflock constellation,” in 28th Annual AIAA/USU Conference on Small Satellites, Logan, Utah, USA, 2014. [3] J. Q. Zhai and X. F. Li,“Introduction of OneWeb system and domestic LEO internet satellite system,” Space Electronic Technology, vol. 14, no. 6, pp. 1–7, 2017. [4] S. Burleigh, A. Hooke, L. Torgerson et al.,“Delay-tolerant networking: an approach to interplanetary internet,” IEEE Communications Magazine, vol. 41, no. 6, pp. 128–136, 2003. [5] K. Fall,“A delay-tolerant network architecture for challenged internets,” Computer Communication Review, vol. 33, no. 4, pp. 27–34, 2003. [6] A. Vahdat and D. Becker, Epidemic routing for partially connected ad hoc networks, Technical Report CS-200006, Duke University, 2000. [7] A. Lindgren, A. Doria, and O. Schelen,“Probabilistic routing in intermittently connected networks,” ACM SIGMOBILE Mobile Computing Communication Review, vol. 7, no. 3, pp. 19-20, 2003. [8] J. A. Davis, A. H. Fagg, and B. N. Levine,“Wearable computers as packet transport mechanisms in highly-partitioned ad-hoc networks,” in Proceedings Fifth International Symposium on Wearable Computers, Zurich, Switzerland, 2001. [9] K. A. Harras, K. C. Almeroth, and E. M. Belding-Royer,“Delay tolerant mobile networks (DTMNS): controlledflooding in sparse mobile networks,” in NETWORKING 2005. Networking Technologies, Services, and Protocols; Performance of Computer and Communication Networks; Mobile and Wireless Communications Systems. NETWORKING 2005, R. Boutaba, K. Almeroth, R. Puigjaner, S. Shen, and J. P. Black, Eds., vol. 3462 of Lecture Notes in Computer Science, Springer, Berlin, Heidelberg, 2004. [10] H. Dubois-Ferriere, M. Grossglauser, and M. Vetterli,“Age matters: efficient route discovery in mobile ad hoc networks using encounter ages,” in Proceedings of the 4th ACM international symposium on Mobile ad hoc networking & computing - MobiHoc '03, Annapolis, Maryland, USA, 2003. [11] K. Tan, Z. Qian, and W. Zhu,“Shortest path routing in partially connected ad hoc networks,” in GLOBECOM '03. IEEE Global Telecommunications Conference (IEEE Cat. No.03CH37489), San Francisco, CA, USA, 2003. [12] S. Merugu and E. W. Zegura, Routing in space and time in networks with predictable mobility, Technical Report, GIT-CC04-07, Georgia Tech College of Computing, 2004. [13] D. Fischer, D. Basin, and T. Engel,“Topology dynamics and routing for predictable mobile networks,” in 2008 IEEE International Conference on Network Protocols, Orlando, FL, USA, 2008. [14] S. Iranmanesh and K. W. Chin,“A novel mobility-based routing protocol for semi-predictable disruption tolerant networks,” International Journal of Wireless Information Networks, vol. 22, no. 2, pp. 138–146, 2015. [15] C. Liu and J. Wu,“Routing in a cyclic MobiSpace,” in Proceedings of the 9th ACM international symposium on Mobile ad hoc networking and computing - MobiHoc '08, Hong Kong SAR, China, 2008. [16] M. Huang, S. Chen, Z. Ying, and Y. Wang,“Cost-efficient topology design problem in time-evolving delay-tolerant networks,” in 2010 IEEE Global Telecommunications Conference GLOBECOM 2010, Miami, FL, USA, 2010. [17] M. Huang, S. Chen, Y. Zhu, and Y. Wang,“Topology control for time-evolving and predictable delay-tolerant networks,” IEEE Transactions on Computers, vol. 62, no. 11, pp. 2308–2321, 2013. [18] H. C. Yan, J. Guo, and H. J. Zhang,“Performance evaluation of routing algorithms on space delay/disruption tolerant networks,” Chinese Space Science and Technology, vol. 36, no. 4, pp. 38–46, 2016. [19] Y. Wang, B. Liu, W. Yu, and B. Zhao,“Routing algorithm for navigation constellation based on evolving graph model,” Chinese Space Science and Technology, vol. 32, no. 5, pp. 76–83, 2012. [20] S. Jain, K. Fall, and R. Patra,“Routing in a delay tolerant network,” in Proceedings of the ACM SIGCOMM, Portland, Oregon, USA, 2004. 10 Space: Science & Technology [21] J. Ott, D. Kutscher, and C. Dwertmann,“Integrating DTN and MANET routing,”in Proceedings of the 2006 SIGCOMM workshop on Challenged networks - CHANTS '06, Pisa, Italy, 2006. [22] D. Yu and Y. Ko,“FFRDV: fastest-ferry routing in DTN-enabled vehicular ad hoc networks,” in 2009 11th International Conference on Advanced Communication Technology, Gangwon, Korea (South), 2009. [23] E. P. C. Jones, L. Li, J. K. Schmidtke, and P. Ward,“Practical routing in delay-tolerant networks,” IEEE Transaction on Mobile Computing, vol. 6, no. 8, pp. 943–959, 2007. [24] J. Leguay, T. Friedman, and V. Conan,“Evaluating mobility pattern space routing for DTNs,” in Proceedings IEEE INFOCOM 2006. 25TH IEEE International Conference on Computer Communications, Barcelona, Spain, 2005. [25] S. Iranmanesh, R. Raad, and K. W. Chin,“A novel destination-based routing protocol (DBRP) in DTNs,” in 2012 International Symposium on Communications and Information Technologies (ISCIT), Gold Coast, QLD, Australia, 2012. [26] J. Wu and Y. Wang,“A non-replication multicasting scheme in delay tolerant networks,” in The 7th IEEE International Conference on Mobile Ad-hoc and Sensor Systems (IEEE MASS 2010), San Francisco, CA, USA, 2010. [27] L. Junhai, Y. Danxia, X. Liu, and F. Mingyu,“A survey of multicast routing protocols for mobile ad-hoc networks,” IEEE Communications Surveys & Tutorials, vol. 11, no. 1, pp. 78– 91, 2009. [28] L. Junhai, X. Liu, and Y. Danxia,“Research on multicast routing protocols for mobile ad-hoc networks,” Computer Networks, vol. 52, no. 5, pp. 988–997, 2008. [29] E. W. Dijkstra,“A note on two problems in connexion with graphs,” Numerische Mathematik, vol. 1, no. 1, pp. 269–271, 1959. [30] B. B. Xuan, A. Ferreira, and A. Jarry,“Evolving graphs and least cost journeys in dynamic networks,” in WiOpt'03: Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks, INRIA Sophia-Antipolis, France, 2003. [31] S. Eshghi, M. H. R. Khouzani, S. Sarkar, N. B. Shroff, and S. S. Venkatesh,“Optimal energy-aware epidemic routing in DTNs,” IEEE Transactions on Automatic Control, vol. 60, no. 6, pp. 1554–1569, 2015. [32] P. Mundur, M. Seligman, and G. Lee,“Epidemic routing with immunity in delay tolerant networks,” in MILCOM 2008 - 2008 IEEE Military Communications Conference, San Diego, CA, USA, 2008. [33] E. Bulut and B. K. Szymanski,“Friendship based routing in delay tolerant mobile social networks,” in 2010 IEEE Global Telecommunications Conference GLOBECOM 2010, Miami, FL, USA, 2010. [34] E. Bulut and B. K. Szymanski,“Exploiting friendship relations for efficient routing in mobile social networksExploiting Friendship Relations for Efficient Routing in Mobile Social Networks,” IEEE Transaction on Parallel and Distributed System, vol. 23, no. 12, pp. 2254–2265, 2012. [35] K. Chen and H. Shen,“SMART: lightweight distributed social map based routing in delay tolerant networks,” , Austin, TX, USA, 2012 20th IEEE International Conference on Network Protocols (ICNP), 2012. [36] P. Hui, E. Yoneki, S. Y. Chan, and J. Crowcroft,“Distributed community detection in delay tolerant networks,” in MobiArch '07: Proceedings of 2nd ACM/IEEE international workshop on Mobility in the evolving internet architecture, Kyoto, Japan, 2007. [37] D. E. Sterling and J. E. Hatlelid,“The IRIDIUM system-a revolutionary satellite communications system developed with innovative applications of technology,” in MILCOM 91 - Conference record, McLean, VA, USA, 1991. [38] C. E. Fossa, R. A. Raines, G. H. Gunsch, and M. A. Temple, “An overview of the IRIDIUM (R) low Earth orbit (LEO) satellite system,” in Proceedings of the IEEE 1998 National Aerospace and Electronics Conference. NAECON 1998. Celebrating 50 Years (Cat. No.98CH36185), Dayton, OH, USA, 1998. Space: Science & Technology 11
第40页
30
第41页
31 Mission Design of an Aperture-Synthetic Interferometer System for Space-Based Exoplanet Exploration Feida Jia1,2, Xiangyu Li1,2*, Zhuoxi Huo3 , and Dong Qiao1,2 1 School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081,China. 2 Key Laboratory of Autonomous Navigation and Control for Deep Space Exploration, Ministry of Industry and Information Technology, Beijing 100081, China. 3 Qian Xuesen Laboratory of Space Technology, Beijing 100094, China *Corresponding author. Email: lixiangy@bit.edu.cn Abstract: In recent years, exoplanet detection has become the technological frontier in the field of astronomy, because it provides evidence of the origin of life and the future human habitable exoplanet. Deploying several satellites to form an aperture-synthetic interferometer system in space may help discover “another Earth” via interferometry and mid-infrared broadband spectroscopy. This paper analyzes a space-based exoplanet exploration mission in terms of the scientific background, mission profile, as well as trajectory design, and orbital maintenance. First, the system architecture and working principle of the interferometer system are briefly introduced. Secondly, the mission orbit and corresponding transfer trajectories are discussed. The halo orbit near the Sun-Earth L2 (SEL2) orbit is chosen as the candidate mission orbit. The low-energy transfer via stable invariant manifold with multiple perigees is designed and the proper launch windows are presented. A speed increment less than 10 m/s is imposed for each transfer to achieve the insertion of Halo orbit. Finally, the tangent targeting method (TTM) is applied for high-precision formation maintenance with the whole velocity increments of less than 5×10−4 m/s for each spacecraft when the error bound is 0.1 m. The overall fuel budget during the mission period is evaluated and compared. The design in this paper will provide technical support and reliable reference for future exoplanet exploration missions. 1. Introduction Search for extraterrestrial life, explore “another Earth” is an eternal theme for human and inspires generations of planetary scientists. It not only improves our understanding of the formation and evolution of planets during the formation of a star system, but also helps scientists to investigatethe possible conditions and criteria of the existence of life. The 2019 Nobel Prize in Physics was awarded for the first discovery of an exoplanet near a Sunlike star, in recognition of the work “for contributions to our understanding of the evolution of the universe and Earth’s place in the cosmos.”, one half of which to James Peebles “for theoretical discoveries in physical cosmology,” and the other jointly to Michel Mayor and Didier Queloz “for the discovery of an exoplanet orbiting a solar-type star.” [1]. In the past 30 years, scientists have discovered more than 4,000 exoplanets, but exploration still has a long way to go. Due to the far distance, searching for exoplanets requires high sensitivity and high resolution. The space-based telescope can eliminate the interference of the Earth atmosphere on observations and becomes a trend in exoplanet exploration. Several telescopes have been launched into orbit such as The Kepler telescope and TESS (Transiting Exoplanet Survey
第42页
32 Satellite). But limited by the launch capability, the size of a single space-based telescope cannot be as large as required by explorations further than current missions which employ methods such as transit, radial velocity. As a next-generation telescope configuration, the array telescope is expected to realize the cross-generation of the telescope system so that Earth-like exoplanet near a Sun-like host star could be detected and characterized via direct imaging, provided that technological challenges being overcome, and opens a new era of high-resolution astronomical observation [2]. The array telescope will form a long-baseline interferometer in the space, which largely improves the observation resolution. Several mission concepts of the array telescope for exoplanet detection have been proposed in the past twenty years. For example, the TPF mission [3] and Darwin mission [4] are proposed by NASA and ESA respectively. They both use multiple (3-5) satellites to form a formation of fixed architecture near the halo orbit to search for exoplanet life. NASA’s Stellar Imager (SI) project [5] plans to deploy 20-30 satellites to form formations near the beam axis in Lissajous orbit around Sun-Earth libration point, and each satellite is equipped with a mirror with a diameter of 1 m to perform 0.1 milliarcseconds (mas) spectral imaging of the surface of stars and the entire universe. Encouraged by the significance of exoplanet exploration, in 2019, CASC also proposed a habitable exoplanet exploration mission by array telescope called MEAYIN mission [6]. The essence of the mission in this study is formation flying around libration point. Many researchers devoted to this theme in recent years. The concept of formation flying around equilibrium was first proposed by Barden [7], and he gave the concept of using the central manifold of Halo orbits to realize natural bounded formation. Heriter et al. [8, 9] defined the regions above as low drift regions and expressed them in the form of second-order surfaces, and studied the small-size formations near Halo orbit and the large-size formations on the Lissajous orbit. It is found that the initial spherical surface will become an ellipsoidal one, while it will become smaller as the distance from the Earth decreases. Several studies [10–12] aimed at the design of formation flying to obtain the stable configuration in CRTBP or bi-circular model (BCM). Meanwhile, the solar sail could be applied to achieve a synchronized formation tracking with a virtual leader. With this background, this paper elaborates the mission analysis in terms of the scientific background, mission profile, as well as trajectory design, and orbital maintenance. The structure of the rest of this paper is as follows. In Section 2, the architecture and principles of exoplanet exploration are discussed. The mission constraints and accuracy requirements on formation flight are given. In Section 3, the candidate mission orbit is selected, and the correspond ing transfer trajectory is designed under the ephemeris model. Section 4 deals with the formation configuration maintenance. The high-precision impulsivebased control method of tangent targeting method (TTM) is applied to satisfies the stability region constraints. Finally, section 5 draws the conclusion. 2. Mission Architecture and Principle of Interferometry 2.1. Observation Demand and Architecture of Array Telescope The array telescope mission intends to observe various types of celestial bodies such as extrasolar habitable planets, solar system celestial bodies, protoplanetary disks, and active galactic nuclei. Searching for and characterizing habitable exoplanets in our solar system’s neighbors (within 65 lightyears) places a high demand on observations. Including 1) High spatial resolution. The star-planet angular distance is better than 0.01 arcsec 65 light-years away from the Sun. 2) High contrast. The brightness of planets and stars differ at least by 7 orders of magnitude in the mid-infrared band 3) High sensitivity. The brightness of the planet in the signal dominant band is less than 3 photons/sec/m2 . 4) Wide Spectral range. Indirect observation in the nearinfrared band of 1 to 5 μm and direct observation in the nearinfrared band of 1 to 13 μm. The spatial resolution of the interferometer is related to the accuracy of path difference control, baseline length and integration time. Therefore, the baseline of the interferometer should be designed according to the resolution requirement. Meanwhile, the control accuracy of optical path difference and integration time should match the baseline range. One of the difficulties in direct imaging of extragalactic habitable planets is the high contrast between the brightness of Earth-like planets and their star. It is difficult to suppress the stellar radiation effectively in the mid-infrared band by using a single aperture optical telescope, while zero-order interferometry in multi-satellite formation can theoretically meet the requirements of high contrast and high resolution at the same time. In this project, the array telescope system consists of a collector in the center and four detectors evenly distributed around. The architecture of the on-orbit
第43页
33 observation scene is shown Figure 1. The dash lines between the detectors and the collector represent the optical communication links. The array telescope system includes a detector system and a collector system, both of which are divided into two parts of platform and load. The platform is divided into five function modules: the structure and organization, thermal control, position control, energy and information management. The load of the detector system includes collecting optical system and telescope detectors, while the load of the collector system consists of combining optical system and interferometric measuring device. $UUD\7HOHVFRSH6\VWHP ([WUDWHUUHVWULDO 3ODQHWDU\6\VWHP FIGURE 1: On-orbit observation scene of array telescope system. 2.2. Principle of Interferometry Interference arrays usually aim at the star to be observed when they are in operation. At this point, the star light propagates along the principle optical axis of the array and reaches the telescope, while the planetary light arrives at a small angle θ away from the principle optical axis of the array. For ordinary binary interference systems, the light incident along the principle optical axis propagates out of the beam through the left and right arms of the interference array with the same optical path (or optical path with difference of integer times of the wavelength), so the interference fringes are mutually reinforcing at the zero-order position. The two-element nulling interferometer introduces the phase delay of π (half wavelength) in one arm (as shown in Figure 2), so the interference fringes formed by the light incident along the principle optical axis offset each other at zero-order. There can be mutually reinforcing bright fringes at zero-order of the interference fringes, which are formed by two kinds of light of along and slightly deviated from the principle optical axis. ʌ Ʌ % Ʌ 6WDU 3ODQHW 7HOHVFRSH 7HOHVFRSH &RPELQHU FIGURE 2: Schematic diagram of two-element nulling interferometer [13]
第44页
34 The four-element nulling interferometer is proposed on the basis of the usual two-element nulling interferometer. By introducing the inverse signal with another pair of interference beams, the nulling range is greatly widened, and the nulling depth is improved by more than three orders of magnitude, which provides great help for the detection of exoplanets, especially Earth-like stars. The optical path of the four-element nulling interferometer is shown in Figure 3. Oi (i =1, 2, 3, 4) are the interference lights in four positions.     7HOHVFRSH 7HOHVFRSH 7HOHVFRSH 7HOHVFRSH FIGURE 3: Schematic diagram of four-element nulling interferometer. It combines two two-element nulling interferometers with different spacing (the spacing of the internal and external pair are 2 s and 4 s respectively), and adjusts the beam intensity ratio so that the amplitude of the external pair is exactly half of that of the internal one. The interference amplitudes of these two pairs cross the zero point with the same value but opposite sign, so as to achieve high-order limitation. Based on the characteristics of the observation demand and the principle of interferometry, the general requirements for the array telescope system are shown in Table 1. These requirements will be considered in the following trajectory design and maintenance. TABLE 1: General path dependent technical requirements. Index Value Inertia pointing accuracy≤ 2'' (3σ) Formation baseline range 40 m-300 m Relative position control accuracy≤ 10 cm(3σ) Load platform position accuracy≤ 1 mm(3σ) Main optical axis pointing accuracy≤ 0.1'' (3σ) Transfer and maintenance velocity increment≤ 1.6 km/s 3. Mission Orbit Selection and Transfer Trajectory Design 3.1. Mission Orbit Selection for Exoplanet Exploration The mission orbit is crucial for the exoplanet exploration by array telescope. The ideal mission orbit should keep away from the electromagnetic interference in the vicinity of Earth and operate in a relatively clean dynamic environment to reduce the magnitude and frequency of orbit maintenance. There are two kinds of candidate mission orbits for the mission, the Sun-Earth libration orbit and the Earth trailing orbit. Due to the properties of three-body dynamics, the SunEarth libration orbit can maintain a relatively fixed position with Earth and the Sun, which is convenient for tracking and control. The spacecraft can transfer into orbit via stable invariant manifolds, which requires nearly zero energy
第45页
35 after launch. Meanwhile, the array telescope can stay in the libration orbit for a long time with very little fuel. By choosing the amplitude of the orbit, the array telescope system can avoid eclipse permanently. Therefore, the array telescope system can have a long working life. The Earth trailing orbit is the other ideal mission orbit, which stays close to the Earth, although undergos a slow drift away from the Earth orbit. The Earth trailing orbit can provide full-time lighting conditions and a stable thermal environment. However, the spacecraft on the Earth trailing obit will continuously drift away from the Earth, which stresses the communications system. Moreover, it requires more launch energy to send the space-based array telescope to the mission orbit and the spacecraft must perform correction maneuvers to stay at the desired position. Based on the above reasons, the Sun-Earth libration orbit is chosen as the mission orbit for the space-based array telescope system. There are many kinds of periodic and quasi-periodic orbits near the SEL2 point including planar and vertical Lyapunov orbits, halo orbits, Lissajous orbits, and axial orbits. Planar Lyapunov orbits always move in the XY plane of the Sun-Earth rotating frame, while the vertical orbits have an “8”-shape and move mainly along the Z-axis. Lissajous orbits are non-closed orbits, of which the characteristic frequency in the XY plane and Z direction are independent. The projection of Lissajous orbit in the XY plane is approximately elliptical, but the orbital plane twists in each period. Halo orbits are closed curves in threedimensional space. The in-plane frequency and outplane frequency of halo orbits are coupled. Meanwhile, several quasi-periodic orbits surround the periodic orbits and form two-dimensional torus, which provides more flexible options for mission design. Both the plane and the vertical Lyapunov orbits will suffer from the eclipse, which affects the performance of the telescope system. The Lissajous orbit also passes through the X-axis nearby region in a long run. Moreover, the stability of large amplitude Lissajous orbits is poor. In the meantime, the simulation shows that when the amplitude of the halo orbit is larger than 10,000 km, the spacecraft can avoid the occlusion of the Earth and achieve full-time light. Therefore, the halo orbits are more suitable as the mission orbit for the spacebased array telescope system. A comparative analysis of halo orbit and Earth trailing orbit is given in Table 2. Considering the stability and fuel consumption, the amplitude of halo orbit is selected to be about 150,000 km in the Z direction. The mission orbit in the rotating frame is shown in Figure 4. FIGURE 4: Mission orbit for exoplanet exploration in the rotating frame. TABLE 2: Comparison and analysis of halo and Earth trailing orbit. Index Halo orbit Earth trailing orbit Shadow occlusion No occlusion No occlusion Cost of orbit insertion Relatively low Relatively high Distance to Earth 1.5 million km Stable 1 million to 10 million km Gradually increase Earth communication coverage Always visible Always visible Velocity increment of coverage Relatively small Relatively large Maintenance frequency Once 3-4 months Once half a year
第46页
36 3.2. Transfer Trajectory Design for Exoplanet Exploration Based on the selected periodic orbit, the design of the corresponding transfer trajectory is investigated. The stable invariant manifold of the periodic orbit is used to find the low-energy transfer opportunity. The transfer trajectory is designed as follows. Firstly, based on the circular restricted three-body problem (CRTBP), the stable manifolds of the target mission orbit are generated at different phase angles and the branch approaching the Earth is selected. Then, the Poincare map is selected according to the perigee state constraint. Afterward, the corresponding manifold that satisfies the height constraint of the parking orbit is chosen as the initial guess of the transfer trajectory. Figure 5 shows the chosen halo orbit and its stable manifold extending to perigee truncated by the Poincare map. Since the stable manifold and the halo orbit are designed based on the CRTBP, the nominal transfer trajectory has a large error, when considering the orbital eccentricity of Earth, and the perturbation forces such as the gravitational force of the Moon. Finally, the conventional differential correction algorithm is adopted to modify the transfer trajectory in the high-fidelity model. Considering the weight and size of the satellite, it is difficult to transport five satellites to the mission orbit simultaneously, so three launches are planned to deploy four detectors and one collector to the mission orbit. The collector and one detector are launched for the first time, two detectors for the second time, and the remaining detector for the last time. Since the size of the formation is much smaller than the amplitude of halo orbits, the specific configuration of the formation can be ignored in the transfer orbit design. If we choose the altitude of parking orbit as 200 km, the proper stable manifold in CRTBP corresponds to the phase angle of the halo orbit in intervals [290° 330°] for the first perigee and the transfer time is between 170.2 and 173.5 days. For the second perigee, the phase angle [170° 220°] and the transfer time are about 243 days. It should be noted that 0° refers to the position on the XZ plane whose velocity components along the X and Z axis are zero. Figure 5 presents the stable manifold passing perigee for the first and second time, while the lowest heights of these two families are highlighted with the green and blue thick dashed lines. That means there are several nearly zero-energy transfer opportunities with the help of the stable manifold. The window will expand if a small injection velocity is applied to adjust the perigee altitude. However, some transfers correspond to large speed increments considering the ephemeris, in which case more stable manifolds passing through the Earth’s perigee more times will be considered. FIGURE 5: Stable manifolds extending to perigee of Earth.
第47页
37 In order to enable the four detectors and the collector to form the configuration and implement observation as soon as possible after launch, the three launches are designed to be completed in one orbital period, which can maximize the observation mission time during the lifetime of the satellites. Based on the analysis of CRTBP, record all the perigees of the manifold under the ephemeris model whose transition time is within two halo orbital periods, and select the solution with less cost to adjust the height of the perigee. At the same time, the time interval between each launch is restricted to more than three months considering ground constraints. The parameters of the three launches are presented in Table 3. The interval of each launch is larger than three months, and the whole three velocity increments for insertions are less than 30 m/s. The initial velocity increment from the parking orbit to the stable manifolds is about 3.19 km/s. As the statement above, we choose the stable manifold of the 1st, 4th, 7th perigee pass-through to minimize the insertion cost. The three transfers in the Earth J2000 and the rotating frame are shown in Figure 6. The total transfer time from the first launch to the last insertion is about 543 days, and the total velocity increment can meet the mission requirement. We further change the first insertion time to see the influence of time on the launch state. It is concluded that the time of the mission have few influence on the velocity increment of the transfers, for the initial velocity increments at different times are almost the same, and the insertion velocity increments for the adjustment of perigee height are very small with the order of m/s. (a) (b) FIGURE 6: Transfer trajectories and mission orbit for exoplanet exploration. (a) Earth J2000. (b) Sun-Earth rotating frame. TABLE 3: Parameters of three transfers. Group Launch time Initial ΔV (km/s) Insertion time Insertion ΔV (m/s) Insertion phase (deg) 1 st 2026-12-26 3.1900 2028-05-16 7.60 281 2 nd 2027-06-03 3.1903 2028-06-21 0.023 353 3 rd 2027-12-11 3.1905 2028-06-10 2.20 331 4. Impulsive-based Control for Formation Maintenance Due to the strong nonlinearity of the dynamics near libration points, the formation nearby needs to be maintained to resist divergence. The required accuracy of relative distances between each detector and the collector is given in Table 1 during the mission execution. Therefore, control is required to maintain the normal execution of the observation mission in the natural evolution of configuration when the relative position exceeds the constraint level. Several kinds of control methods are investigated in recent years, such as the LQR method, the improved polynomial eigenstructure assignment (PEA) method [14], a multi-agent, nonlinear,
第48页
38 and constrained optimal control method [15], and geometric control method [16], etc. The above-mentioned continuous control method is applicable near the equilibrium point sensitive to small disturbances. However, in some scenarios, it is necessary to use an impulse control strategy while excluding continuous control to meet specific mission constraints. The aiming method based on the differential correction scheme called the equitime targeting method (ETM) proposed by Howell and Baden [7] is a widely adopted strategy. The nominal path is divided into segments of a given time, and impulse maneuvers are performed with the nominal state at the end of each segment as the target. The fixed time interval leads to waste in some trajectory sections with small errors, and the unnecessary switch on and off will also cause unnecessary engine loss. In 2012, Qi [17] et al. proposed the tangent targeting method (TTM) based on a two-level differential correction, which could fully and efficiently satisfy the predetermined error bound. The comparison between the TTM and ETM shows that the number of maneuvers can be significantly reduced and the length of time between successive maneuvers can be greatly increased. For the maintenance of the formation configuration of the libration point, the constraint of maximum drift error bound exists in this project, so the control law of TTM [17] is more applicable, which can maximize the time spent within the error bound between maneuvers. Figure 7 shows the definition of the square configuration direction in the rotating system and the configuration parameters of the formation. Let n be the normal of the formation, and its direction is defined by two angles α,β in the spherical coordinates. β is the angle between n and axis Oz , and α is the angle between the projection of n on plane Oxy and axis Ox, C is the collector located at the centroid. The globe with the rotating system is used to facilitate the representation direction of the configuration, and the little tiny blue body in the globe represents the Earth. In Figure 7, the positions of the four detectors are noted as r1, r2, r3, r4, and the size of the configuration is defined by the lengths of the four arms (1) At the beginning of the mission, the size of the configuration is given by (2) When the formation propagates on the mission orbit, due to the perturbation, four detectors gradually deviate from their nominal positions relative to the collector, causing the deformation of the configuration. Through previous research [18], it is found that when the initial configuration plane near the libration point points to the X-axis of the rotating system, the stability of the formation is best. Therefore, in this mission, the positions of four detectors relative to the collector in the rotating system are selected to make the configuration plane point to the X-axis of the rotating system, which indicates that (3) For a space-based exoplanet exploration mission, it is required that each detector is located at a fixed position relative to the detector. That is to say, the detectors dynamics are modeled as a perturbation relative to the motion of the collector (4) where δr0 is the desired relative position. / & D E +DORRUELW Q  6  6  6  6 ; < = Q &  6  6  6 6 U U U &ROOHFWRU U 'HWHFWRU (a) (b) FIGURE 7: Square configuration of the formation. (a) Direction definition of the formation. (b) Arms of the formation.
第49页
39 As is shown in Figure 8, the desired path of the detector is divided into segments of a given error bound. Given the start and the end of a segment, noted as (5) where Δt is the time spent of the considered segment. Assume that the relative state of the collector is (6) The desired orbit is satisfied at the start tk of the segment, . According to the Taylor expansion, in segment Equation 5, we have (7) Substitute into Equation 7, and set , then (8) where is the velocity at the start of the segment after impulsive control, . After preliminary analysis, the k-th time span for orbit maintenance can be approximately given by (9) where εmax is the allowed error bound, and then the correction equation after derivation can be formulated as (10) where ε s is the maximum position error in the natural propagating time corresponding to the s-th iteration for k-th orbit maintenance. Here we assume four detectors of 500 kg form a square formation with the size of 200 m, and the halo orbit with an amplitude of 150,000 km is chosen as the mission orbit. The relative positions of the four detectors are (11) The tolerance of position error is set as δrk =0.1 m throughout the whole mission. The control quantity and control deviation of the four detectors are shown in Figure 9. N G  Y N G  Y ' N Y N  G   Y N  G   Y ' N Y N G U N  G  U 2UELWRIFROOHFWRU 'HVLUHGRUELWRIGHWHFWRU 5HDORUELWRIGHWHFWRU FIGURE 8: Principle of tangent targeting method.
百万用户使用云展网进行网络电子书制作,只要您有文档,即可一键上传,自动生成链接和二维码(独立电子书),支持分享到微信和网站!
收藏
转发
下载
免费制作
其他案例
更多案例
免费制作
x
{{item.desc}}
下载
{{item.title}}
{{toast}}