SAR Target Recognition Based on Transfer Learning and Data Augmentation with LSGANs

Abstract

This paper presents a method combining transfer learning and data augmentation based on least squares generative adversarial networks (LSGANs), which can effectively solve the problem of lacking training samples in the synthetic aperture radar (SAR) target recognition algorithm. Compared with the conventional methods of data augmentation, the new generated data are more realistic when using the LSGANs to expand the data set, and the recognition accuracy is significantly improved. Transfer learning can apply characteristics extracted from the source domain for the target domain, which effectively reduce the need for training samples in existing algorithms. Features were extracted by using ResNet50 that has been pre-trained, and convolution neural network (CNN) was used as a classifier for SAR target recognition. Experimental results based on real radar data sets testify that the method proposed in this paper significantly improves the recognition accuracy.