Research On Video SAR Moving Object Tracking Based On Deep Learning Technology

Video SAR has the characteristics of all-time,all-day,high-precision and real-time imaging,which has high military application value.Moving target tracking is the key problem of effective field reconnaissance by video SAR.Video SAR images have the characteristics of complicated background and drastic change of target appearance.Therefore,how to track video SAR moving target stably and efficiently is a very challenging task.In this paper,deep learning method is adopted to conduct in-depth research on video SAR moving target tracking.The main work is as follows:1.To solve the problem of similar distractors in the process of target tracking,we design the Instance-Aware Backbone network.In-depth analysis of the root cause of the inability of target tracking models to effectively distinguish similar interferers,and on this basis,the Instance Aware Backbone network(IB)is designed.IB can effectively distinguish different instances of the same object,which can largely avoid the drift phenomenon of tracker caused by similar interferers.2.Aiming at the drastic change of target shape during the tracking process,a template image update strategy based on Projected Gradient descent(PGD)is proposed.The imaging characteristics of video SAR lead to dramatic changes in the apparent morphology of moving targets.Therefore,it does not make sense to continue using the first frame target image as a trace template.PGD algorithm can update the template image effectively and avoid the failure of tracking task due to the change of target shape.3.For the practical application of end-to-end tracking of video SAR targets,a video SAR target tracking model is constructed.Based on Siam RPN++ tracking framework,a video SAR target tracking network is constructed by the Instance-Aware Backbone network and the template image updating strategy of PGD.Experiments show that the network can achieve stable tracking of moving targets in video SAR.In this paper,a large number of comparative experiments are carried out on video SAR target tracking datasets,and the experimental results prove that the proposed algorithm can effectively improve the success rate and accuracy of tracking compared with the benchmark algorithm.When tested on video SAR target tracking dataset,IB target tracking algorithm improves the success rate by 2.2% and the precision by 2.1%.The target tracking algorithm using PGD template updating strategy improves the success rate and precision by 0.6%.Video SAR target tracking improved by 3.6% in success rate and 3.1% in precision.