Study on SAR Target Recognition Performance Based on Convolution Neural Network

Though recent some methods of Synthetic Aperture Radar(SAR) image target recognition based on convolutional neural networks(CNNs) have been improving the recognition accuracy extremely, the procedures and guidance involved in the network design still require further research. Therefore, it is desirable to explore the influence of each element on network for achieving a higher recognition performance. In an effort to find out the role of different types network layers, we adopt the LenNet for the basic model and improved it. Moreover, we modify the network layer with some compare experiment to obtain highest accuracy while preserving its time complexity basically unchanged. Use this way to justify that our network(B3 network) have a outstanding recognition accuracy(99.3%) on the public data set.