Abstract: Coronavirus 2019 (COVID-19) is a highly infectious and fatal virus that originated in Wuhan, China, and has since spread globally. To categorize X-ray pictures into the three categories of normal, pneumonia, and COVID-19, we trained several deep convolutional networks with two open- source datasets using the suggested training procedures. 15664 X-ray pictures of COVID-19-infected patients were processed utilising a variety of methods for optimal results. In this paper, we propose new training approaches to support the network in learning when the dataset is imbalanced, with a disproportionate number of instances of COVID-19 and a greater number of cases from other classes. It is suggested to integrate the Xception and ResNet50V2 networks into a neural network. This network was able to attain the maximum degree of precision achievable by merging data from two robust networks. For this study, we used 11,302 images that reflect the accuracy that can be attained in real-world scenarios. The proposed network achieves a 99.50% accuracy rate in identifying COVID-19 cases.