Face mask detection and classification for COVID-19 using deep learning

Kayali, D.; Dimililer, K.; Sekeroglu, B.

With the emergence of COVID-19, our lifestyles have changed. Because of its high infectivity and high severity, fighting with this virus has become a global problem in no time. Minimizing the speed of spread and keeping the numbers under control is a way to save more people by taking care of them better and also have more time spent on research to find a cure such as medicine or vaccine which will put an end to this situation. During this time, personal protection like using face masks is very important since it protects people and others close to them and significantly reduces the risk of infection on correct usage. Unfortunately, some people can act careless or reckless which puts many people at risk. This leads us to use automated systems in crowded places to detect those who do not follow the rules. When it comes to automated systems, artificial intelligence-related applications are favorable for supporting humans. In this paper, at first, a dataset was obtained by adding face masks to the existing Labeled Faces in the Wild (LFW) dataset to detect three mask-wearing conditions;correct, wrong and no mask. Then NASNetMobile and ResNet50 networks were trained using the considered dataset. The ResNet50 outperformed the NASNetMobile by achieving 92% detection accuracy. © 2021 IEEE.