Double Face Masks Detection Using Region-Based Convolutional Neural Network

Abstract

Because of the fast spread of coronavirus, the globe is facing a significant health disaster of COVID-19. The World Health Organization (WHO) released many suggestions to combat the spread of coronavirus. Wearing a face mask in public places and congested locations is one of the most effective preventive practices against COVID-19. However, according to recent research wearing double face masker even provide better protection than just one mask. Based on this finding, various public places require double masks to proceed more. It is pretty tricky to monitor individuals in crowded public places personally. Therefore, a deep learning model is suggested in this paper to automate recognizing persons who are not wearing double face masks. A faster region-based convolutional neural network model is developed using the picture augmentation approach and deep transfer learning to increase overall performance. We apply deep transfer learning by fine-tuning the low level pre-trained Visual Geometry Group (VGG) Face2 model. This study used the publicly accessible VGGFace2 dataset and the self-processed dataset. The findings in this study show that deep transfer learning and image augmentation can increase detection accuracy by up to 11%. Consequently, the created model achieves 93.48% accuracy and 93.19% F1 score on the validation dataset, demonstrating its excellent performance. The test results show the proposed model for further research by adding the predicted dataset and class.