Video SAR Target Detection and Tracking Method Based on Yolov5+Bytetrack

Video Synthetic Aperture Radar (SAR) technology can acquire high frame rate image sequences of the observed scene, and use the shadow features formed by ground moving targets such as vehicles in the image sequence to achieve dynamic target state perception. In this paper, a lightweight Yolov5s deep learning network is used for vehicle target shadow detection, and image enhancement and attention mechanisms are introduced to significantly improve the detection accuracy of vehicle targets, achieving an accuracy rate of 95.6% and greatly reducing the false alarm rate. Furthermore, five target tracking algorithms are compared, and it is verified that the Bytetrack algorithm can achieve better dynamic target data association and tracking performance than SORT and its three improved methods.