SAR Target Recognition with Generative Adversarial Network (GAN)-based Data Augmentation

In convolutional neural network (CNN)-based synthetic aperture radar (SAR) target recognition, a large number of SAR image datasets are required to train the CNN. To deal with the problem of insufficient training datasets, a generative adversarial network (GAN) can be applied for data augmentation. In this paper, the performance of SAR target recognition with GAN-based data augmentation is investigated using the MSTAR datasets. The results show that, by applying the GAN-based data augmentation using a proper data augmentation factor, a target recognition accuracy as high as 99.38% can be achieved, which is improved by 2.35% compared to the condition without data augmentation. Thus, the GAN-based data augmentation provides a promising solution to enhancing the accuracy of SAR target recognition.