Thin and Thick Cloud Removal on Remote Sensing Image by Conditional Generative Adversarial Network
Xiaoke Wang, Guangluan Xu, Yan Wang, Daoyu Lin, Peiguang Li, Xiujing Lin · 2019
Cloud removal is an essential step to enhance the quality of cloud-covered remote sensing image. In recent years, conditional Generative Adversarial Network (cGAN) yields promising improvement in plentiful image-to-image translation tasks. In this paper, we propose a novel objective function to upgrade the structural similarity index based on cGAN. We discover that ImageGAN is effective to focus on global information for thick cloud-covered images and Patch-GAN has fewer parameters while maintaining outstanding performance for thin cloud-covered remote sensing images in the experiments. Experimental results demonstrate that our method achieves remarkable performance in both PSNR, SSIM and visual effect on cloud-covered remote sensing images especially thin cloud-covered images.