Abstract
Due to the spread of the COVID-19 cases, medical experts are analyzing the automatic detection model for preventing the disease. In worldwide, there is more number of confirmed cases, but only a few CT scan images are available. The classes like COVID-19, normal and pneumonia images are collected from various benchmark databases. This work has the stages like pre-processing, data augmentation, feature extraction and classification. Initially, the image contrast is enhanced and Gaussian filtering is used for eliminating the noise. Then, the image augmentation techniques like image rotation; scaling, translation and flipping are applied on the original dataset. Then, the two stage hybrid deep learning (DL) model has been designed for the diagnosis of COVID-19 using the CXR images. The DL model hybrid ResNet with YOLO-V5 classifier is used for extracting the features and classifying the COVID-19 classes. Further, for optimizing the YOLO-V5 classifier, the met heuristic algorithm Capuchin Search Optimization (CaSO). The proposed ResNet with YOLOV5-CaSO automatically classifies the normal, COVID-19 and Pneumonia cases. The outcomes provided by this model proved that this model can efficiently diagnose COVID-19 and Pneumonia using the CXR images. Further, the accuracy and TPR achieved are 99.3% and 1.108 for the proposed model and it shows that this model saves time and assists the early diagnosis.
Keywords: Capuchin Search Optimization, COVID-19, CXR Images, Deep Learning, Image Augmentation.