Covid-19 Detection Based on Chest X-ray Images Using Attention Mechanism Modules and Weight Uncertainty in Bayesian Neural Networks

The novel of Coronavirus (Covid-19) has effective in worldwide year by year, which led to a serious problem for affected lungs with different sequelae from patient. The techniques in Vision has a higher performance for Covid-19 detection, using deep learning methods to extract features, specifically Convolutional Neural Networks (CNN) can get more information for prediction. To mitigate more higher complexity feature representation subspaces, attention mechanism can obtain significant feature for convolution operation. In this study, the proposed method used Convolutional Block Attention Module (CBAM), Vision Transformer (ViT) and Swin Transformer (SwinT), also integrate the Bayesian neural network (BNN) that it can be optimized with Weight Uncertainty can achieve the generalization ability, the experimental results show that it has more robustness and higher performance than other methods. The experimental shows that Densely Connected Convolutional Networks (DenseNet) combined with ViT and BNN have better performance than other methods.