Abstract:  Cloud segmentation and detection of remote sensing images plays a pivotal role in the field of weather forecast. Many meteorologic applications such as precipitation forecast, and extreme weather forecast, depend on the results of cloud detection. In this paper, based on the satellite remote sensing image dataset, we propose a CNN based algorithm to address this cloud segmentation problem, which can achieve pixel-level cloud segmentation results on high resolution, large scale, multi-channel satellite images. The proposed algorithm consists of three steps: 1) image patching; 2) patch image segmentation; 3) image stitching, which makes cloud segmentation in the pixellevel precision. In comparison with the traditional methods, it has the advantages of independence of expert knowledge, totally data motivated approach, and more robustness in extreme cases. It is demonstrated by the experimental results that our proposed algorithm can satisfy the requirements of the weather forecast, and thus has a strong potential to be put into business usage.