Abstract: Channel attention mechanism can effectively utilize different feature channels to improve the classification ability of convolutional neural networks. However, for the method using global average pooling to obtain the global features of channels, there is a high probability that different channels in the feature graph have the same mean, which makes the features after global average pooling lack diversity, and further affects the classification performance of the network. To solve this problem, a shred channel attention mechanism is proposed to construct a module, which extends the output dimension of global average pooling, enhances the feature diversity of global average pooling layer in channel attention, and then uses one-dimensional convolution to calculate channel attention. Image classification experiments are performed on CIFAR-100 and ImageNet datasets by combining the split-channel attention mechanism with multiple residual networks. The experimental results show that the segmentation channel attention mechanism can effectively improve the accuracy of the model, and it also shows better advantages compared with other attention mechanisms.