Research On Shadow-based SAR Multi-target Tracking Method

Synthetic Aperture Radar(SAR)is not restricted by weather and light,and can even complete high-resolution imaging tasks in severe weather and night.The continuous development of SAR imaging technology drives the continuous improvement of imaging resolution,and the system of imaging algorithms is constantly innovating to adapt to different scenarios.The research on the detection and tracking of video SAR moving targets has also made rapid progress,and has been widely used in the fields of video supervision and rescue.With the rapid development of deep learning,the application of deep learning to SAR moving target detection,tracking and imaging has also achieved a lot of results.This paper applies deep learning to SAR shadow tracking tasks,and combines traditional imaging algorithms to build a SAR ground moving target tracking and refocusing framework.In the framework built,first obtain SAR shadow video through SAR target simulation and video SAR imaging system,then use tracking algorithm to track the shadow of the ground moving target in the SAR video,and finally reconstruct the target trajectory according to the tracking result to complete the moving target refocus imaging.Specifically,the main work and innovations of this article are as follows:1.SAR ground moving target tracking and refocusing framework construction and simulation verification.The built SAR ground moving target tracking and refocusing framework uses the video SAR back projection algorithm to obtain SAR shadow video,uses the multi-target tracking algorithm to obtain the moving target trajectory,and combined with the tracked trajectory and echo data,the moving target is refocused and imaged through the moving target back projection algorithm.Using the built framework,the moving target imaging azimuth offset analysis,azimuth defocus qualitative analysis and SAR shadow characteristics are compared and verified by experiments.2.Production of video SAR moving target simulation data set and measured data set.Using the video SAR backward projection algorithm in the framework,adjust the parameters to simulate the SAR shadow video of the road scene,and annotate the obtained simulation video and the measured video published by San Diego to make the data set of the subsequent multi-target tracking algorithm.3.Comparison of the multi-target tracking algorithm based on background subtraction and the Anchor-based multi-target tracking algorithm.If the multi-target tracking algorithm based on background subtraction is used for SAR shadow video tracking,background jitter and detection target holes will occur,which is not suitable for SAR shadow video tracking.The Anchor-based multi-target tracking algorithm based on deep learning is used in SAR shadow video tracking,the tracking results are analyzed,and the moving target back projection algorithm is used for refocusing imaging.Despite the defocus,the moving target can be obtained roughly imaging.4.Propose an improved Anchor-free multi-target tracking algorithm and perform refocus imaging.Based on the existing Anchor-free multi-target tracking algorithm,the improvement of the SAR shadow video tracking algorithm is carried out from three aspects: the introduction of attention mechanism,the modification of the loss function and the optimization of the network architecture.First of all,the embedding branch is retained,the attention mechanism is introduced into the algorithm to improve the tracking effect,and the loss of the embedding branch is changed to triple loss,which is more suitable for multi-target tracking of SAR shadow video than the cross-entropy loss,and further improves the tracking effect.Then,considering that the appearance characteristics of SAR shadow video in the overall algorithm are weak,in order to simplify the network,an Anchor-free multi-target tracking algorithm is proposed that removes the embedding branch and introduces attention mechanism,which greatly simplifies the network model and significantly reduces the amount of calculation.While speeding up the running speed,it achieves the tracking effect equivalent to that of retaining the embedding branch.Finally,combined with the built SAR ground moving target tracking and refocusing framework and the proposed Anchor-free multi-target tracking algorithm,the refocusing imaging of the SAR moving target is carried out,which improves the imaging effect of the SAR moving target.