SAR image to optical image translation technology based on conditional generative adversarial network

Abstract

To address the problems of poor readability and difficult interpretation caused by the special imaging mechanism of Synthetic Aperture Radar (SAR) images, this paper combines the latest advances in Generative Adversarial Network (GAN) technology in machine learning to overcome the problems of CycleGAN In this paper, we combine the latest advances in GAN technology to overcome the problems of unstable training, failure to converge, and lack of diversity in generating a single image, and construct a supporting training dataset to design and optimize a multimodal image translation network model to explore a solution for translating SAR images into easily understood optical images. The research results of this paper are very important for realizing applications such as alignment, matching and change detection between multimodal images.