Study Of SAR Image Target Detection Based On Deep Network

The target detection of Synthetic Aperture Radar(SAR)is a comprehensive interdiscipline which includes signal processing,radar imaging,image processing,artificial and intelligence,and it is widely used in military and civil fields.Due to the complexity,diversity and variability of the SAR image,the application of SAR image target detection in practical enviroment is still in an immature stage.Therefore,SAR image target detection is a challenging research subject.Firstly,this thesis summarizes the research background and the development trend of SAR target detection briefly,and then introduces the main research work and the content arrangement.Secondly,the important part of this thesis focuses on the traditional methods of SAR target detection and the latest research of deep network.Finally this thesis introduces the SAR image target detection in the complex scene based on deep network.The main work of this thesis is as follows:1.The method of SAR image target detection which based on CFAR is studied in the thesis.Firstly,the hypothesis of SAR target detection is mathematically modeled by the description of the SAR target detection problem,and then the maximum likelihood estimation method is used to deal with the parameters of the four basic CFAR models.Finally,the adaptive scenes and merits and demerits of the four basic CFAR detectors are analyzed experimentally.2.The method of SAR image target detection which based on deep learning is studied in the thesis.With the recent research of state-of-the-art target detection based on deep learning,the SAR image target detection based on YOLO and Faster-RCNN are proposed in this thesis.Compared with the traditional CFAR target detection,these methods can achieve end-to-end target detection,and their performance is better than traditional methods.In addition,the SAR image target detection of GLOB-YOLO is proposed in this thesis which can achieve real-time processing in small SAR scene.3.An online learning idea for SAR image target detection is proposed to solve the problem of the shortage of training data.With using the new data on the line to constantly update the deep networks of SAR target detection proposed in the thesis,the problem of the shortage of training data can be converted into a cycle process of online training which is detecting while updating.Finally,the online cycle system of SAR image target detection is achieved by using Qt Creator,which can bring great convenience for the following research work.