SAR Target Classification Method Based on Convolutional Neural Network

This paper constructs a characterization network that describes the characteristics of SAR target images. In the feature space extracted by the feature description network, class centers are established according to the training data. To recognize the test image samples, the characterization network transforms the test samples into the feature space of the class centers, and the similarity between the test sample and three classes are calculated. The proposed method does not rely on professional knowledge to describe the SAR target characteristics. It only utilizes data-driven method to obtain better recognition results, which can reduce the time consumption for modeling the recognition target. The proposed method achieves a classification accuracy of 99.65% on three classes of MSTAR dataset. And 100% classification performance was achieved on certain class.