Face Mask-Wearing Classification Using Transfer Learning Technique with MobileNet V2

Abstract

The COVID-19 pandemic has adversely affected the global economy, politics, society, and other areas. It has dramatically led to a loss of human lives worldwide and is continuing to have a heavy impact on people's health and living conditions. In order to avoid the spread of this pandemic, mask-wearing is required by law or regulation in most places, especially indoors public places. Using Deep Learning techniques to detect and authenticate people wearing masks which can help recognize patterns or behavior of the public, contributing to limiting the rapid spread of COVID-19 pandemic is becoming effective, beneficial, and widespread. However, prior research studies only involve differentiating between the use of a face mask or not. In spite of the fact that wearing face masks incorrectly will not protect ourselves, especially children, from virus transmission and will reduce the effectiveness of COVID-19 prevention, a limited study has been done to solve the problem of mask-wearing. We use the technique of transfer learning with MobileNet V2 to train the model from the Face Mask Label Dataset (FMLD) and Flickr Faces HQ (FFHQ) dataset to not only detect if a mask is used or not, but also classify the status of mask wearing over the faces. The results show that the trainning accuracy of experiment model is 99%.