Evaluating deep transfer learning models for face mask detection

Goh, Pei Jin (2022) Evaluating deep transfer learning models for face mask detection. Final Year Project, UTAR.

Preview

PDF Download (3185Kb) | Preview

Abstract

Due to the fast transmission of coronavirus and the severe sequela of COVID-19, which has no specific cure, the world is facing a massive health crisis. According to the World Health Organization (WHO), wearing a mask in public locations and crowded locations is the most effective prevention of COVID-19. In Malaysia, wearing a face mask is mandatory in public areas. However, it is impossible to detect all passers-by manually as it requires much manpower. This research proposes an automation approach to maskwearing detection by identifying people who are (i) not wearing a mask, (ii) wearing a mask, (ii) incorrect mask-wearing, and (ii) wearing double masks. Transfer learning methods were adopted by using five pre-trained models: (i) VGG, (ii) MobileNet, (iii) ResNet, (iv) Inception and (v) Xception models. These models were trained based on 2000 real-life data sets collected from various sources with a data augmentation technique. The research results show that the pre-trained ResNet152 model outperformed the other models by achieving 0.8667 accuracy on the testing data set (120 images from the other distribution) and 0.8447 accuracy on the videos captured using a smartphone.

Actions (login required)