Auxiliary Diagnosis of Pneumonia Based on Convolutional Attention and Parameter Migration

Abstract

Traditional deep learning models have limited ability to extract features from pneumonia images. This study combines convolutional attention modules with transfer learning to improve the model's feature extraction ability and simplify training.This article has discovered an improved CBAM-Xception neural network pneumonia model. This network uses Xception as the main network, and combines with the convolutional attention CBAM module to enhance the expression of lesion information and suppress irrelevant information interference; At the same time, transfer learning is introduced to prevent over fitting when the sample data amount is small. In order to evaluate the effectiveness of the optimization model, experimental simulation tests were conducted on the Mendeley Data public pneumonia dataset. The improved model achieves 94.2% accuracy in classifying four images of COVID-19, bacterial pneumonia, viral pneumonia and normal chest on the test set, and the optimization effect is significant. In order to further verify the performance of this method, the experiment divided the small sample data set to train the model, and divided the large sample data to test the generalization performance of the model. The results show that the model in this paper has good generalization ability. This model can provide an important basis for the auxiliary diagnosis and treatment of pneumonia.