MULTI-VIEW CNN-LSTM NEURAL NETWORK FOR SAR AUTOMATIC TARGET RECOGNITION

Synthetic aperture radar (SAR) has always received wide attention for its developing performance in military and civil applications. SAR automatic target recognition (ATR) is an important research field of the SAR application with the growing number and resolution of the SAR images. SAR images will be greatly influenced by the imaging azimuth, which could also be utilized to extract the correlation features between the adjacent azimuths. In this paper, we proposed a multi-view convolutional neural network and long short term memory (CNN-LSTM) network to extract and fuse the feature extracted from different adjacent azimuths. It adopts the structure of convolutional neural network to extract the optimal feature from the SAR images. Then, the structure of multiple layers of the long short term memory is adopted to fuse the optimal features of adjacent azimuths. Finally, a softmax is employed as the classifier to get the recognition results. Experimental results based on the MSTAR data set have shown the effectiveness and accuracy of the proposed method.