Abstract: To address the problem that physician subjective factors can affect the Computed Tomography images diagnostic accuracy of COVID-19 (novel coronavirus infection) and CAP (Community-Acquired Pneumonia) types, a MobileNetV2-SELN structure is proposed based on the improved MobileNetV2 network. First, the Block module in the MobileNetV2 structure is improved by adding SE blocks and scale attention mechanism, and a fully connected layer and global pooling is introduced to facilitate the acquisition of multi-scale features. Then, for the characteristics of large similarity between COVID-19 images and CAP images, GroupNorm is used instead of BatchNorm2d to enable the model to better acquire pneumonia features. Finally, the model is optimized using SGD optimizer. The experimental results show that the classification accuracy of the model proposed in this paper has a higher classification accuracy.