一种超轻量化卫星目标形心定位算法  

Lightweight centroid locating method for the satellite target

作　　者：刘路远 韩璐瑶 李娇娇[1] 夏晖[2] 饶鹏[2] 宋锐[1] LIU Luyuan;HAN Luyao;LI Jiaojiao;XIA Hui;RAO Peng;SONG Rui(State Key Laboratory of Integrated Services Networks,Xidian University,Xi’an 710071,China;CAS Key Laboratory of Infrared System Detection and Imaging Technology,Shanghai Institute of Technical Physics of the Chinese Academy of Sciences,Shanghai 200083,China)

机构地区：[1]西安电子科技大学空天地一体化综合业务网全国重点实验室,陕西西安710071 [2]中国科学院上海技术物理研究所,中国科学院红外探测与成像技术重点实验室,上海200083

出　　处：《西安电子科技大学学报》2023年第6期120-132,共13页Journal of Xidian University

基　　金：芜湖-西电产学研合作专项资金(XWYCXY-012021002);国家自然科学基金(61901343);111计划(B08038);地理信息工程国家重点实验室开放基金(SKLGIE2020-M-3-1);中国博士后科学基金(2018T111019);陕西高校青年创新团队。

摘　　要：天基光电探测单元具有视场角大、载重小和机动灵活的特性,在空间作业的监视任务中有重要应用,其中卫星目标的识别和形心定位是其重要功能之一。由于目前在轨载荷处理器的算力较弱,在近距离测角任务中无法使用深度神经网络算法进行复杂定位。针对这一问题,深入分析了目标的特点,依据卫星几何特性设计了一种超轻量化实时处理方法。该算法设计了直线特征和帆板边缘特征提取器,提出了一种利用特殊几何关键点计算目标最小外接矩形的策略,提升外接矩形与目标边缘的贴合度。算法在仿真数据集上进行了测试,并在物理模拟试验场采集的影像数据上进行了验证。测试结果表明,算法的检测精度优于主流深度学习目标检测网络Yolov5n,计算量仅为对比算法的10%。在嵌入式实时处理平台上移植后,实时定位速度可达到5 120×5 120@5 fps。在验证环境下,基于所提算法形心定位的结果进行测角,测角精度优于0.05°,满足应用场景的精度要求。The space-based photoelectric detection unit is vital in satellite identification and positioning.It has a large field of view,a small load,and flexible maneuvering properties.However,the computational capability of the CPU mounted on an on-orbit satellite could be much higher,which can hardly afford the necessity of deep learning networks.In this paper,we analyze the character of space targets deeply and designed a lightweight real-time processing algorithm.We specifically design feature extractors for the line and satellite contour patterns in the algorithm and propose a minimum bounding box calculation strategy.The algorithm is tested on the simulation dataset and verified on the images captured by the physical emulation platform.Testing results demonstrate the effectiveness of our algorithm.The detecting accuracy of our algorithm is better than YOLOv5n,and the computational load is only 10%that of the competitive methods.We transplant the algorithm to an on-orbit real-time processing platform.The positioning speed reaches 5120×[e-mail] accuracy for angle measurement based on the centroid positioning results is more significant than 0.05°,which meets the requirement for real application systems.

关 键 词：空间站 卫星 目标定位 目标检测 实时处理 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]