Research On Moving Target Detection Method Of Video SAR Based On Shadow

As a new type of radar system in the field of Synthetic Aperture Radar,Video SAR can realize round-the-clock,all-weather,low delay and uninterrupted monitoring of the target area with its high frame rate and high resolution imaging characteristics,which provides a new way for timely detection and real-time tracking of important targets especially moving targets in modern military and civil fields.On the Video SAR image,the moving target will form a shadow visible to the naked eye in the real position,which can continuously and intuitively reflect the current position and state of the targets.In contrast,the traditional use of echo characteristics in moving target detection is inevitably restricted by the "blind speed".Therefore,the method of moving target detection using shadow is favored by many scholars,and currently,the method of moving target detection based on shadow has become a research hotspot in many applications of Video SAR.However,compared with the optical field,there is still a big gap in the research of Video SAR moving target detection.It is a topic of great research value to further reduce false alarms and missed detections and improve the detection rate of moving targets.Based on this,the moving target detection of Video SAR based on shadows is studied in this thesis.The main work is as follows:1.The working principle and imaging characteristics of Video SAR system are studied.According to the formation mechanism and characteristics of target shadow,the feasibility of shadow-based detection algorithm is analyzed,which provides a strong theoretical support for follow-up shadow-based methods.2.Aiming at the problems of low shadow quality and poor contrast in traditional image processing methods,a method of using adaptive threshold segmentation and image fusion enhancement strategy to enhance moving target shadow is proposed.After the experimental verification on the real data of Video SAR,the detection effect of moving target shadow is improved to some extent,and false alarms and missed detections are suppressed.3.In view of the vacancy of Video SAR label data set and the lack of research on deep learning methods,the Video SAR data released by Sandia National Laboratory in the United States are tagged accurately,and the end-to-end detection of Video SAR moving target is realized on this data set.By means of Auto-anchor algorithm,selecting and improving feature extraction network(Backbone)and optimizing training methods,a moving target detection model based on improved Faster RCNN Video SAR is proposed,and the corresponding evaluation indices are established to evaluate the performance of the model.The experimental results show that the model has good robustness.Among them,the research on lightweight network of backbone also provides a basis for the application of detection method based on deep learning to Video SAR aircraft platform.4.To solve the problem of insufficient shallow information in Video SAR images,a Video SAR moving target detection method based on information fusion is proposed.The image fusion method is used in the preprocessing of the Faster RCNN detector,and the feature pyramid network is used in the neck connection,which enhance the model’s acquisition of target shadow and multi-scale information.The effectiveness of the proposed method is verified by experiments.