Abstract: For traditional multi-label image classification models, it is difficult to generate high-level image features that are closer to related labels, and the visual correlation between the labels is not used, which leads to problems such as insufficient recognition accuracy. A multi-label image classification algorithm based on spatial attention and graph convolutionis proposed in this paper. Firstly, the graph convolutional network is used to learn the features of the label adjacency graph and the GLOVE algorithm is usedto obtain the label embedding from the label sequence. Secondly, an improved spatial attention networkis introducedin the high-level semantic information to re-calibrate the semantic features of a specific category and suppress background and interference information.Finally, the high-level semantic information with the tags extracted by the graph convolutional network in the classifier based on co-occurrence feature fusionare integrated, and the final prediction of the modelis completed in the channel one-to-one method. Experiments on two public data sets show that the average accuracy of the proposedalgorithmon the MS-COCO and VOC-2007 data sets are 81.42% and 94.3%, which are 1.13 and 1.3 percentage points higher than the basic MLGCN. The amount of model parameters is only one-eighth of the original model, and the number of iterations required in the training process is far less than that of the original model, which greatly reduces its training cost.