COVID-19 MEDICAL FACE MASK CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORK TO DETECT PROPER AND IMPROPER WEARING OF FACE MASK

Abstract

Since the emergence of COVID-19, the wearing of a face mask has been seen to be an effective method of limiting the transmission of the virus, with some governments making it compulsory, especially in public spaces. Proper way of wearing a face mask necessitates covering both the nose and mouth completely. However, some are still resistance of wearing it; due to inconvenience or discomfort, whilst some are simply wearing it incorrectly, which in turn, reduces the efficacy of the mask and in some circumstances, making it useless. This paper provides a method of determining whether a face mask is being properly or improperly worn using Convolutional Neural Network (CNN) via images. The proposed method uses a large MaskedFace-Net dataset, containing labelled images of correctly and incorrectly masked faces, for both training and testing. ResNet and YOLO models are used to detect faces from images, which then identify the wearing state of the mask. It has been shown that both methods are able to attain classification accuracies of above 90% using the dataset. The result is significant, as the face mask classification can be used for automatic monitoring of face masks in real-time in public places such as hospitals, airports and crowded premises, to ensure compliance to the current COVID-19 guidelines.