Improved Classification of CT images of Cooropneumonia Based on ResNet50

In the CT image diagnosis of COVID-19, use traditional convolutional neural network such as ResNet to extract the feature of target will lose the information between the channel and the space, and when performing convolutional subsampling, more details of the image will be lost. In order to overcome the defect, we propose an improved ResNet50 network model, which introduces a CBAM attention mechanism after each residual module of ResNet50. In the fourth and fifth parts of ResNet50, convolutions with a convolution core size of 3 are transformed into empty convolutions. When training data, a transfer learning method based on fine tuning is used. The experiment shows that the improved ResNet50 on the COVID-19 CT dataset have high recognition accuracy than AlexNet, VGG16, and GoogleNet. Accuracy of improved ResNet50 model is 98.901%, which is 6.593 and 1.099 percentage points than original model and transfer learning model. Compared with the traditional convolutional neural network, the improved model proposed in this paper has greatly improved the recognition accuracy of COVID-19.