ISAR Images Generation Via Generative Adversarial Networks
Ruoyi Zhou, Zhilong Yang, Feng Wang
Abstract
One of the challenges faced by current intelligent target recognition tasks is the lack of samples, especially in the Inverse Synthetic Aperture Radar (ISAR) images understanding. In this paper, we proposed an ISAR objects generative network to generate multi-aspect ISAR images. A simulated ISAR dataset of six types of aircrafts is produced via, using bidirectional analytic ray tracing (BART) method. Then, the proposed generative network is trained with the simulated ISAR dataset. We evaluated the performance of the proposed network using structural similarity (SSIM). The experimental results show that the generated targets are very close to the real ISAR samples, and the SSIM between generated and real ISAR images of aircrafts is larger than 0.7.