Research On Synthetic Aperture Radar Image Target Detection And Recognition Method

Synthetic Aperture Radar(SAR)is a high resolution coherent imaging radar.Compared to infrared,visible and other technologies,SAR can operate around the clock and in all weather.Traditional SAR image target recognition methods rely on the distinguishability of manually designed features.For complex SAR images,traditional techniques lose a large amount of image information in the image pre-processing and feature extraction process,and the whole process is repetitive and complex,which has more shortcomings.Deep learning can automatically obtain and mine the key features of different objects to improve the accuracy of object detection and recognition.To this end,this paper conducts a correlation study of SAR image target detection and recognition based on deep learning,and its specific content and research focus include:(1)For the problems of convolutional neural networks(CNN),such as insufficient feature information extraction of SAR images,false targets caused by complex background interference,and low detection performance,a non-local channel attention network(NLCANet)is proposed.First,based on Goog Le Net structure combined with an asymmetric pyramid non-local block(APNB),we can capture more context information and enhance the correlation between pixels and regions.Second,the squeeze-and-excitation block(SEB)is added to the Inception structure to become Inception-SEB(ISEB).The model can obtain channel dependence based on the fusion of features at different scales through ISEB.Finally,the experimental results based on the moving and stationary target acquisition and recognition(MSTAR)and SAR ship detection dataset(SSDD)show that the proposed method improves the detection ability of targets in complex backgrounds and achieves better land and sea target recognition performances.(2)In order to solve the problems of low detection accuracy and a high number of missed ship targets in high-resolution SAR,we propose an improved YOLOV5 algorithm.First,the small target ships are located more accurately by increasing the size of the input SAR images and optimizing the anchor frame of the ship targets.Then,in order to reduce the interference of coastal backgrounds and obtain accurate positioning of ship targets,APNB and Sim AM are introduced.Then,the C3 output is channelmixed to enhance the information flow between channels,so that the output has more and richer feature information.Then,to reduce the number of parameters and computational cost during model training,the normal convolution in the NECK part is replaced with Ghost convolution.Finally,the superiority of the improved method over other advanced methods is verified on the High-Resolution SAR images datase(HRSID)and SSDD,and the robustness of the improved method was verified on the AIR-SARShip-1 dataset.