COVID-19 Image Classification Based On Deep Learning

The novel coronavirus(SARS-Cov-2)poses a serious threat to human health,infects a large number of patients and threatens life and health.Stopping the spread of the virus promptly and isolating patients infected with COVID-19 can greatly reduce the number of patients.The most critical step in containing the spread of COVID-19 is the timely and effective treatment of patients with suspected infection,but reverse transcription polymerase chain reaction(RT-PCR)testing for patients is relatively timeconsuming and has a high false-negative rate,which is not conducive to control the outbreak.To further detect suspected patients and improve the accuracy of detection methods,this paper uses artificial intelligence algorithms to classify lung medical images based on deep learning models to detect infected patients,and the main work is as follows:1.A multi-channel dual-attention network(MDA-Net)based on deep transfer learning is proposed to detect lung images.Firstly,a multi-channel dual-attention module is introduced in the framework of deep transfer learning,which uses the position relationship of multiple channels to fuse image features at different scales.Then,the attention mechanism and lightweight convolutional neural network are combined,which expands the MDA-Net receptive field and improves the feature extraction ability of complex regions and marginal regions of lung images.Finally,the performance of MDA-Net is verified on different datasets.It is shown that MDA-Net achieved an average accuracy of 99.25% and 99.39% in the binary classification task and 3-classification task respectively,showing good classification performance.2.An active attention convolutional neural network(AMC-Net)based on transfer learning is proposed to classify lung images.Firstly,a new active attention fusion(AEA)module is introduced to extract channel attention,which uses the active attention mechanism to extract feature information more efficiently.Secondly,AMC-Net is combined with a model pre-trained with a large amount of data and transferred learning,which solves the problem of lack of lung medical image data and imbalance of data types by transfer learning,and accelerates the convergence speed of the model.Finally,by visualizing the image prediction process and classification results,the model decision-making process is made clearer.In the task of classifying lung X-ray images,AMC-Net achieved an average accuracy of 99.36%.In the binary classification task of lung CT images,AMC-Net achieved an average accuracy of 100.00%,and in the 3-classification task,AMC-Net achieved an average accuracy of 99.93%.Through a series of ablation experimental data comparison,the effectiveness of AMC-Net is proved.Compared with the existing classification models,it is proved that AMC-Net has strong competitiveness.