Cloud Detection via Convolutional Neural Network in Visible Light Remote Sensing Images

Abstract

Cloud detection in visible light remote sensing images is a challenge due to only few RGB spectral channels and multiple types of clouds and land covers. In this paper, we propose a new method based on convolutional neural network for cloud detection. Our method automatically learns the intrinsic features of the clouds and the other land covers, and can produce high accuracy for cloud detection. The learned features in our method are then visualized and explained in detail, and compared with other features in a baseline method. Finally, experiments are conducted to demonstrate the effectiveness of our method.