Eff2CaNet:Convolutional Neural Networks Based on Coordinated Attention for COVID-19 Classification

Medical image classification based on deep learning has become an important diagnostic tool. Since the outbreak of the novel coronavirus pneumonia, serious harm to human health. In the face of a large number of COVID-19 patients, doctors rely on their personal theories and experience to diagnose lung CT, which is prone to diagnosis errors. Therefore, the construction of accurate and fast auxiliary diagnosis system can effectively improve the diagnosis efficiency. An improved Eff2CaNet architecture based on the EfficientnetV2 network is proposed. The model is built on the basis of the network EfficientnetV2, replaces the SE module in the EfficientnetV2 architecture using Coordination Attention (CA), and effectively captures the relationship between the location information and the channels, thus accurately obtaining the areas of interest and improving classification accuracy. The accuracy of the model proposed in this paper on the COVID-19 CT data set was 93.1%.