Design And Realization Of Vision Detection Algorithm For Deep Space Target

While people are exploring and using the space field,the development of space technology has also ushered in new opportunities and challenges,such as the establishment and maintenance of space stations,the recovery,release,and maintenance of satellites.In addition,after experiencing frequent space activities for nearly a century,the space is full of large amounts of space junk,which poses a greater potential threat to future space exploration.In addition,the aerospace-grade CPU used in spacecraft has a harsh working environment,facing the challenges of cosmic radiation and a temperature difference of more than 300 degrees Celsius.Therefore,the development of aerospace-grade CPU is slow.Nowadays,the computing power of aerospace-grade CPUs used at home and abroad is generally low,so how to ensure In the case of detection accuracy,the realization of lightweight object detection algorithms has become the focus of the on-orbit service.Therefore,this paper takes the tracking of space non-cooperative targets as the background and takes space non-cooperative satellites as the target,and designs a lightweight and high-precision object detection algorithm.First of all,in response to the serious shortage of spacecraft computing resources,we made lightweight improvements to the YOLOV5 algorithm,using the Mobile Net V1 network that combines the Res Net and CSPNet structure as the backbone network,and using deep separable convolution to replace the traditional convolution in the detection head.The YOLOV5＿DW＿RES＿CSP model ensures the accuracy of model detection while the model size and calculation amount are reduced by half.Secondly,in order to verify the feasibility of the proposed algorithm in the field of space on-orbit service,we propose a six-degree-of-freedom UR5 visual servo system based on the YOLOV5＿DW＿RES＿CSP model and the KCF object tracking algorithm.The system is divided into three modules: object detection,object tracking,and robotic arm tracking.The object detection module algorithm completes the preliminary detection of target location information after the deep space target data set we constructed is trained.The object tracking module is based on the initial target location information.Tracking is performed to obtain the real-time position information of the target,and the robot arm tracking module uses the real-time position information of the target as input to plan the path and realize the motion control of the robot arm.Finally,a ground simulation environment was built to complete the construction of a complete mechanical arm visual servo system,and the system implemented was verified by experiments.The experimental results show that the performance of each module of the system meets the standard,and the overall operation of the system is smooth.The visual servo system realized by this subject with the addition of object detection and object tracking is more intelligent and more efficient.