A Hybrid Approach for predicting COVID19 using Multiple Convolution Neural Networks and Self Attention Maps

Abstract

Covid-19, the most infectious ailment effected due to severe acute respiratory syndrome, which has hindered the health of the people worldwide by causing severe respiratory problems and also lead to extent of death. This infectious syndrome needs to be monitored and detected at right time to prevent the growth of Covid-19 pandemic so as to cure the disease through an accurate diagnosis and proper medication. To address this current issue, a Convolution neural network model (CNN) integrated to self-attention has been proposed. The convolution operator is limited to local receptive field being the disadvantage of CNN. So, we have incorporated self attention mechanism between image representations at deep layers so that the model could learn both local and long range dependencies of the image. Therefore, efficacy of the proposed model has been illustrated through the experimental results and had proven to be progressive in detecting Covid-19 infection by equipping the self-attention module to CNN architecture.