Efficient Cloud Detection for CX-6(02) Satellite Images Based on Neural Network

Remote sensing images are widely used in many aspects, such as agriculture, meteorology, military, and navigation. Clouds often appear in remote sensing images, which will affect the quality of the image and influence the subsequent process. Cloud detection has always been a hotspot in remote sensing image processing. There are many kinds of clouds, the variety of brightness, texture, and shape of which bring a lot of inconvenience to automatic detection. In this paper, we design a cloud detection algorithm for CX-6(02) satellite images according to the characteristics of the actual remote sensing image. First, the image is divided into small image blocks by superpixel segmentation, and then the feature of the superpixel block are extracted through a neural network. Finally, a logistic classifier is applied to determine the superpixel block is cloud or non-cloud. Experiments show effectiveness of the proposed algorithm.