A SAR Denoising Network Based on Generative Adversarial Learning

Synthetic Aperture Radar (SAR) images are critical in microwave remote sensing and are used for resource exploration and target location. However, due to the principle of coherent imaging, SAR images are often interfered by speckle noise, which reduces image quality. It is necessary to remove noise but retain details. We are committed to improving the network structure to increase the diversity and authenticity of generated images. Inspired by Se-Res-Unet, we fuse generative adversarial networks for SAR image denoising and named the model SAR-RsuGan. In it, the generator and the discriminator compete with each other to generate high-quality images. We use synthetic SAR images for training and compare with other methods to obtain the best performance.