 This paper proposes an improved cloud classification method based on a densely connected hybrid convolutional neural network, 3D, CNN. This network uses the features of both the spatial and spectral channels of the FY, for a satellite to classify clouds into seven different categories. The proposed network was tested against the cloud-sat 2B, cold-class product and achieved an overall accuracy of 95.2%, outperforming other deep learning-based techniques. This article was authored by Bo Wang, Ming Wei Zhou, Wei Qing, and others.