Cloud detection and estimation for satellite optical images

Abstract

Cloud is a very common weather phenomenon in the world. In Singapore, it is possible that everyone will see the cloud every day. It is not bad to have the cloud over the top. However, for optical satellite images analysis, which operates tens of thousands optical images daily, cloud causes a lot of trouble. Therefore, cloud detection and cloud masking process become an important procedure in data preprocessing for optical satellite images. The problem is that the existing algorithms such as Sentinel-2 are too complicated and or too costly. They need satellite using Multi-spectral camera to record the image daily or they need to analyze the image pixel by pixel which is not so efficient as expectation. Meanwhile, the computer is easy to be confused by different white objects on the land such as snow, white houses. Some of the existing technique using rectangle to segment the cloud area from the image is useful but the shape of the cloud is random and irregular which need more effective method to crop the cloud out. Our method aims to develop and apply some algorithms, with superpixels technique and machine learning, for effectively detecting the cloud on RGB images and evaluating application