Cloud removal in optical remote sensing imagery based on SAR-guided inpainting with artificial intelligence

Abstract

Cloud is a very common weather phenomenon. For optical satellite photo, the occlusion of the image by the cloud layer is very serious. Sometimes cloud will cover up more than 50% of the optical image. Some areas are even unobservable for two to three months. Synthetic aperture radar (SAR) is a modern high-resolution microwave imaging radar. It is able to penetrate the clouds to get information of the ground. However, since SAR is not optical imaging, the imaging effect is very different from the optical image. Computer Version has made a great progress in recent years. Recently, the ability of recognizing processing images is comparable to that of humans. It has achieved automatic extraction, analysis and understanding of useful information from the images. In this project, I will use image processing, SAR image processing and such deep learning tools as Generative Adversarial Network (GAN) and Convolutional Neural Network (CNN) to recover the cloud masked optical image with SAR image.