Research On Multiview SAR Target Recognition Methods

Synthetic aperture radar automatic target recognition(SAR ATR)has the capabilities to locate the regions of interest containing the potential targets from the large-scale-scene SAR images,assign the classified attributes for every SAR target accurately,and transform the image data into the intelligence information,which is currently the frontend hotspot of international SAR technology.Limited by the imaging mechanism,the classical SAR ATR technologies based on single-view pattern are difficult to meet the accurate cognition of the targets.SAR images from multiview observation can provide more complete target information than from a single-view one.And multiview SAR ATR will be beneficial to the accurate target cognition and recognition,which is an important development direction in SAR ATR.How to realize the target detection,feature extraction,classification and flight path optimization are the core and key issues in multiview SAR ATR.This dissertation focuses on these problems,and develops theoretical analysis,method research and simulation verification for multiview SAR ATR.The main innovations are as follows.1.A multiview SAR processing method for simultaneous target detection and image formation is proposed,which can solve the contradiction between the time resource utilization of the SAR system and the detection performance.The sub-aperture SAR images with resolutions from low to high are iteratively detected and accumulated during SAR imagery formation,then the high-resolution detected target regions are obtained.The proposed method can effectively get the high-resolution SAR image and improve the detection rate.2.A fused feature extraction method for multiview SAR targets is proposed,which can comprehensively and balancedly extract the classification feature from the targets.The spatial relationships between the neighborhoods are established,and the optimal projection for feature extraction is obtained,which can extract both the global and local classification features from the multiview SAR images and improve the recognition rate.3.A multiview SAR target recognition method based on deep neural network with a parallel network topology is proposed,which can effectively train the neural networks with limited raw SAR images.The proposed networks can learn the classification information from the SAR images after the multiview augmentation processing,and improve the performance of the multiview SAR ATR.4.A optimal flight path planning method for multiview SAR is proposed.By solving the constrained multiobjective optimization,this method can find out the efficient and safe flight paths for the SAR platform and obtain the corresponding optimal imaging viewpoints,which can further enhance the efficiency and accuracy of the multiview SAR ATR.The methods mentioned above have been verified by the experiments.The results show that the proposed methods can effectively solve the main problems in multiview SAR ATR,and realize the target detection and recognition effectively and accurately.