Abstract: The all-sky camera data is a common method for real-time cloud information monitoring, and cloud volume is one of the first considerations during astronomical site selection. Therefore, automatic classification of the all-sky foundation cloud image based on image quality, application background and other factors, and achieving an automatic classification algorithm with high robustness and adaptability, will provide important help for astronomical site selection. The paper uses the Xuelong all-sky camera data to train the convolutional neural network model, and uses the all-sky camera data of the Lijiang Observatory to test. It has achieved good application results and realized an automated classification method for all-sky camera data with high mobility.