Super-resolution SAR Image Reconstruction via Generative Adversarial Network

Abstract

In this work, we presents a super-resolution (SR) reconstruction method for the synthetic aperture radar (SAR) images based on the generative adversarial network (GAN), SRGAN for short. In comparison with conventional SR algorithms developed in the area of image processing, the proposed SRGAN technique could make an important breakthrough in terms of reconstruction accuracy and computational efficiency for the SAR image SR. To achieve high-resolution, high fidelity and optics photo-like SAR images, SRGAN explores a perceptual loss function consisting of an adversarial loss and a content loss. Selected experimental results based on Terra-SAR datasets are provided to demonstrate the state-of-the-art performance of our proposed method.