Cloud Removal on Satellite Image using Transfer Learning based Generative Adversarial Network

Abstract

Satellite Image Processing (SIP) is important in both academic and practical aspects because it has a wide range of applications. However, since the collected Optical Remote Sensing images often contain cloudy images, it is difficult to extract complete information from satellite image data. Therefore, cloud removal from satellite imagery is important to rethink the efficiency of satellite image processing. Therefore, in this study, we propose a methodology for pre-learning the Generator based on U-net and applying Generative Adverserial Network to the satellite image data of (Cloudy, non-Cloudy) pairs collected based on Google Earth engine. This solves the quantitative problem of data that it is not easy to obtain a usable data set due to weather problems, and it will show that the pre-learning results learned from abundant data in the SIP field are effective.