Classification of pneumonia based on DenseNet and transfer learning

Pneumonia, as an acute respiratory infection, is the most significant infectious cause of children’s death in the world because it will age the lungs. In recent years, it is still one of the most deadly diseases in the world. With the development of artificial intelligence technology, deep learning technology is widely used in the fields of pneumonia detection and classification. Compared with traditional machine learning technology, it is helpless due to data distribution differences, outdated labeling data, and other issues. Transfer learning can label data or knowledge structures from related fields, thus reducing the need for labeling a large number of medical data. On this basis, this paper puts forward a transfer learning method about DenseNet and classifies pneumonia in chest X-ray images by comparing it with ResNet18, MobileNetV2, and DenseNet121. In addition, through the comparative experiment of DenseNet with different depths on the COVID-19 Radiography Database data set, it is verified that the deeper the depth, the stronger the feature extraction ability and better capture the higher-order features. Our model has the best performance in classifying pneumonia after training on the same data set. The outstanding feature of this study is that DenseNet’s transfer learning is used for pneumonia classification, which is suitable for today’s pneumonia classification task.