ABSTRACT
COVID-19, a highly insidious infectious disease, is now widespread in the world and has caused hundreds of millions of dollars of damage. Rapid detection for COVID-19 is essential for outbreak control, yet existing RT-PCR assays require hours to obtain results and may be incorrect. To enable faster and more accurate detection of COVID-19 infected patients, training neural networks on chest X-ray scans and using the trained models to assist in the diagnosis of lung disease is worth considering. Our team constructed a model consisting of 5 CNN networks of AlexNet, VGG11, GoogleNet, ResNet18 and DenseNet121 with a fully connected neural network using transfer learning and ensemble learning. This model is able to combine the advantages of the 5 CNN networks to get better results. At the same time, we use CLAHE image enhancement algorithm with image augmentation to optimize the training set, which avoids overfitting problem and can further improve the results. With the above approach, we can train a highly accurate ensemble model in a short time to quickly detect COVID-19 infected patients with a small sample of chest X-ray images. Our ensemble model converges quickly and the final test accuracy is 96.48%, which is higher than the test accuracy of any of the five individual CNN networks. © 2022 IEEE.