COVID-19 Classification by Transfer Learning and Deep Stacking Technique

COVID-19 Classification by Transfer Learning and Deep Stacking Technique

In recent years, extensive research has delved into the realm of rural The coronavirus disease, commonly referred to as COVID-19, has spread worldwide. Combatting COVID-19 requires thorough screening of infected patients, but most methods involve physical contact which poses, a risk of spreading the disease to doctors and medical practitioners. In this work, the main focus is on rapidly detecting COVID-19. Herein, Transfer learning technique is adopted to initialize the model parameters, and five pre-trained deep convolutional networks such as Resnet101V2, Resnet152V2, InceptionResnetV2, Xception, and Densenet201 are employed. Further, Softmax is applied for classification of fully connected network, and deep stack is utilized to ensemble the top three-layer. Experimentation performed on the dataset consists of Chest Radiography (CXR) images, which consist of mainly 3 classes: COVID positive, Normal and Pneumonia. The model accuracy of 96.06% is achieved for the proposed fusion pre-trained model which is better in association with the other deep models. Moreover, individual model accuracy of 97.50% is reported by Xception model for the same.