The invention discloses a satellite cloud amount computing method based on a convolution neural network. The satellite cloud amount computing method comprises the following steps: firstly, establishing a satellite cloud picture training sample including 6000-8000 training samples, manually labeling a thick cloud segment, a thin cloud segment and a clear sky cloud block, each of which includes 2000-3000 samples in the satellite cloud picture and taking the segments as a training sample of the convolution neural network; then, pre-processing the training sample and the satellite cloud picture as data input of the convolution neural network, performing convolution neural network detection to detect locations of the thick cloud area, the thin cloud area and the clear sky area in the cloud picture; and finally, according to the locations of the thick cloud, the thin cloud and the clear sky in the cloud picture, respectively computing grey values of the thick cloud, the thin cloud and the clear sky, and performing cloud amount computing of the satellite cloud picture according to the grey values. According to the satellite cloud amount computing method disclosed by the invention, the image of the satellite cloud picture can be directly used as input of CNN (cable news network), and a feature extraction function is integrated into the neutral network for implicitly extracting the features of the image, so that the satellite cloud amount calculating method is relatively convenient and precise in comparison with that of the prior art, and has an important application value.