Abstract: A simple and efficient class-specific residual attention (CSRA) module is proposed to solve the problem that the image features obtained by global max pooling in ML-GCN lack pertinence in different image regions for specific categories and lose image local feature information. . This module can effectively capture different spatial regions occupied by different classes of objects. Furthermore, combining the proposed class-specific residual attention with graph convolutional neural networks, a multi-label image classification algorithm (ML-CSRA) based on multi-head class-specific residual attention and graph convolution is proposed. First, the general image feature map is extracted by convolutional neural network, and then the proposed class-specific residual attention is extended to the multi-head form, and it is applied to the general image feature map extracted by the convolutional neural network. Regions correspond to different categories of features to be extracted. Finally, the label-related features extracted by the graph convolutional neural network are combined with the image features extracted by the multi-head class-specific residual attention to obtain the final multi-label image classification result. The experimental results on MS-COCO 2014 and VOC-2007 datasets show that the proposed algorithm outperforms existing algorithms on all evaluation metrics.