Fighting against COVID-19: A novel deep learning model based on YOLO-v2 with ResNet-50 for medical face mask detection

Abstract

Deep learning has shown tremendous potential in many real-life applications in different domains. One of these potentials is object detection. Recent object detection which is based on deep learning models has achieved promising results concerning the finding of an object in images. The objective of this paper is to annotate and localize the medical face mask objects in real-life images. Wearing a medical face mask in public areas, protect people from COVID-19 transmission among them. The proposed model consists of two components. The first component is designed for the feature extraction process based on the ResNet-50 deep transfer learning model. While the second component is designed for the detection of medical face masks based on YOLO v2. Two medical face masks datasets have been combined in one dataset to be investigated through this research. To improve the object detection process, mean IoU has been used to estimate the best number of anchor boxes. The achieved results concluded that the adam optimizer achieved the highest average precision percentage of 81% as a detector. Finally, a comparative result with related work has been presented at the end of the research. The proposed detector achieved higher accuracy and precision than the related work.