SAR Target Detection And Discrimination Based On Deep Networks

Synthetic Aperture Radar(SAR)is a high resolution radar that is not affected by illumination,and can detect target under all weather,all time conditions and from long distance.So SAR is widely used in every field.The automatic target recognition(ATR)technology of SAR image can extract effective information about target.As the important parts of the SAR ATR system,SAR target detection and discrimination are to extract the location information of the target,eliminating the clutter's influences.But when the scene of the SAR images is complex,it is difficult to distinguish targets from clutters using traditional methods,and the performance needs to be further improved.In recent years,deep learning methods have achieved rapid progress in optical image processing field,but developed slowly in SAR image processing field.So,it is an irresistible trend to improve the SAR image processing methods based on the application of deep learning methods in optical images.This thesis studies the SAR target detection and discrimination algorithms by combining the deep learning theory.The main contents of this paper are summarized as follows:1.The first part presents the background,significance and development of SAR image target detection and discrimination in domestic and abroad.Then,the main work of the thesis are summarized.2.The second part in this thesis mainly studies the SAR target detection based on two-parameter CFAR(Constant False Alarm Rate)algorithm and the SAR discrimination based on traditional method.Firstly,two-parameter CFAR algorithm with the modified background windows is used to improve the performance of the detection in dense target area.Then,the chip getting from the detection stage is discriminated using different features,and the performance and shortcomings are analyed.3.A SAR image target discrimination method based on Convolutional Neural Network(CNN)is studied in this part.Firstly,the basic principles about CNN are introduced.Then,four CNN models with different structures are constructed to discriminate the chips extracted from the detection stage.A set of parameters measuring the performance of the SAR image discrimination are used to discribe the results more accurately.Finally,Non-maximum Suppression(NMS)method is used to optimize the final results.4.This section studies the SAR target detection algorithm using the deep learning method based on candidate regions.Firstly,the mainstream methods of target detection using deep learning method in the field of optical image processing are introduced.Then the algorithm of deep learning method based on candidate regions are expounded.Finally,by analyzing the difference between optical image target detection and SAR image target detection,an effective training data construction method is used,and the faster R-CNN is modified which improves the SAR target detection performance.