A recognition method of COVID-19 CT image based on ResNet network

In order to improve the diagnosis rate of COVID-19 patients, according to the analysis for CT image characteristics, a method based on ResNet for CT images to identify COVID-19 patients was proposed. Combining the ResNet50 network model, by adding channel attention mechanism, adding Dropout function, and embedding the Adam optimizer of cosine annealing method, the average recognition accuracy in this method can reach to 95% for the analysis of confusion matrix results, with high accuracy and low recall rate. The results show that ResNet50 network model with Grad-CAM function has high recognition accuracy for the COVID-19 CT images. Therefore, the automatic recognition method for COVID-19 CT images has a practical application value.