Research On Moving Target Detection Method Of VideoSAR Based On Shadow

VideoSynthetic Aperture Radar(VideoSAR)is a new SAR imaging mode with the characteristic of high frame rate imaging.Its imaging results are presented in the form of video,which can continuously record the changes of the target area and realize dynamic observation,providing a new solution for moving target detection,so it has important research significance.Shadow,as an important feature of moving targets,can reflect the real position and state information of moving targets.The characteristics of VideoSAR high frame rate continuous frame imaging make it possible to detect moving targets by continuous shadow of moving targets.However,the change of scene and motion state of moving target will lead to small Contrast Ration of a Distributed Target(DTCR)and large Extension of Shadow(SE)in VideoSAR image,which leads to increasing the difficulty of moving target shadow detection.Therefore,it is significant to study the theory and method of moving target detection based on shadows in VideoSAR.In this paper,deep learning method is used to extract and enhance the features of weak shadow targets,and realize the shadow detection of moving targets with small DTCR and large SE,which greatly improves the detection performance.The specific work and progress are as follows:1.The principle of VideoSAR imaging is introduced,and the causes of moving target shadow are analyzed.According to the mathematical expression of moving target shadow area and shape,the moving target shadow in each frame of VideoSAR and the interference factors that may affect the detection of moving target shadow with different DTCR and SE are explained.The influences of DTCR,speed and direction of velocity on moving target shadow area are quantitatively studied and simulated,which provides a theoretical basis for VideoSAR moving target detection by using shadow.2.Aiming at the enhancement and extraction of moving target shadow features in VideoSAR images,a moving target detection method based on FPN-DLR is proposed.In this method,Faster R-CNN is constructed by Feature Pyramid Network(FPN)which deeply fuses features by adding points and pixels bit by bit,and the traditional regression is replaced by Dense Local Regression(DLR).The effectiveness of FPN-DLR for VideoSAR moving target detection is proved by simulation and measured data.The results of experiments show that FPN-DLR greatly improves the accuracy of VideoSAR moving target detection compared with the traditional single-frame moving target detection in VideoSAR based on OTSU.3.To solve the insufficient ability of FPN-DLR to detect moving target shadows with large SE and small DTCR,the multi-frame detection method,memory enhanced global-local aggregation(MEGA),is applied to VideoSAR field for the first time and I-MEGA is proposed by improving MEGA.The influence of different layers of residual network on moving target detection in VideoSAR is analyzed.I-MEGA utilizes the correlation between moving target shadows in adjacent frames of VideoSAR to realize feature aggregation and extraction of moving target shadow.Experiments with simulated and measured data show that compared with FPN-DLR and multi-frame detection based on flow-guided feature aggregation,I-MEGA greatly reduces the false alarm error rate compared with FPN-DLR,and effectively improves the ability of detecting moving target shadows in VideoSAR with small DTCR and large SE.