Cloud Detection Algorithm Using Advanced Fully Convolutional Neural Networks in FY3D-MERSI Imagery

摘要

Cloud detection plays a very important role in the development of satellite remote sensing products and influences the accuracy of satellite products that characterize the properties of clouds, aerosols, trace gases, and ground surface parameters. However, current existing cloud detection methods rely heavily on the data of visible bands. It makes FY3D MERSI, which lacks visible band data at night, difficult to use these methods with high accuracy. In this paper, we proposed a cloud detection method based on deep learning termed CM-CNN for FY-3D MERSI. In order to ensure the effect of the network, the data has been strictly selected and consequently preprocessed. The method can automatically extract identified target features and fuse multi-level feature information, and adjust the parameters in the network without setting a threshold. Besides, this method proves to be better and more robust while only using mid-infrared and long-infrared band data in different cases.