Research On Ship Target Recognition In SAR Image Based On Deep Network

The ship target recognition in Synthetic Aperture Radar(SAR)image is an important part of maritime target interpretation,and it is also a hot topic of research at present.Based on the special advantages of SAR image in all-day and all-weather,as well as the urgent needs of SAR image target recognition in military and civil fields,the research on ship target recognition method has high theoretical value and practical significance.The traditional ship target recognition method in SAR image includes four independent modules: detection,identification,feature extraction and recognition.The modules are interdependent,so they have great limitations.Deep learning has attracted much attention since it came out.Due to deep learning outstanding feature extraction ability and selflearning ability,it has been widely applied in the field of image processing.SAR image target interpretation method based on deep network has been extensively studied by scholars in recent years.This thesis mainly studies the ship target recognition method in SAR image based on deep network.The specific work are as follows:Firstly,in this thesis,the background and significance of the research on target recognition in SAR images are introduced,and the research status at home and abroad is reviewed.Then,this thesis focuses on the traditional methods and deep learning methods commonly used in target recognition of SAR images,and gives a detailed theoretical introduction to some key technologies.At the end of this part,the application of deep learning in the field of ship target recognition in SAR image is briefly introduced.Secondly,this thesis studies the class imbalance in ship target recognition,and then from the point of view of mathematics,we to explain the essence of categories imbalances and the influence of the model.Then,this thesis analyzes the advantages and disadvantages of the existing methods to deal with the class imbalance,a new loss function is proposed to deal with the class imbalance in the ship recognition.The loss function optimizes the classification boundary by directly increasing the intra-class distance and indirectly increasing the inter-class distance.At the end of this section,the effectiveness of the proposed method is verified by a large number of experiments.Finally,the integration of detection and recognition are emphatically studied.The integration of detection and recognition are a trend of target recognition in SAR images.However,ship target detection and recognition based on deep network cannot be integrated due to the influence of data sets and methods.In this thesis,the existing data sets are sorted out,and then a five-class ship identification data set is constructed.In order to realize integration and better feature fusion,a deep Network based on pyramid structure is constructed.At the same time,aiming at the problem of high similarity of different types of ships,adding attention model to the pyramid structure can effectively amplify useful features and suppress useless features.The effectiveness of the proposed integration method based on deep network is verified through experiments.Meanwhile,the comparative experiments show that adding attention model can effectively improve the rate of ship recognition.