Generation of SAR Images with Features for Target Recognition

Abstract

Since it is difficult to obtain a large number of the real samples of SAR images, the accuracy of synthetic aperture radar automatic target recognition (SAR-ATR) based on deep learning is often affected by the lack of real samples. Generative adversarial network (GAN) is a method that can effectively generate samples to expand dataset. This paper proposes a GAN that adds a condition to guide image generation and modifies the true and false discriminator to a discriminator with classification (DwC). In addition to correctly recognize the real SAR images, DwC recognizes the generated images as the class N + 1. In order to make the generated images recognized as the real images by DwC, the conditional generator gradually learns to generate the images with features of a specific category. Applying the SAR images generated by our model to target recognition based on deep learning can effectively improve the accuracy.