Adaptive Ensemble Approaches for Face Mask Detection Using Transfer Learning Model

Amidst the Covid-19 pandemic, wearing a mask has become mandatory and remains crucial in the ongoing pandemic. In certain situations and conditions, masks may be required periodically. However, manually monitoring large crowds of people to ensure they are wearing masks in public places or events can be an arduous task that is both time-consuming and difficult to manage. One way to address the challenge of manually monitoring mask-wearing in public places is to adopt machine learning (ML) methods and ICT-based surveillance mechanisms, thereby reducing the need for large numbers of personnel. While previous studies have proposed various machine learning models for mask detection, there is still a possibility for enhancement in their accuracy. Additionally, these models can have lengthy training times due to the standard 80:20 train and test dataset ratios used. This study suggests an ML-based face detection that utilizes MobileNet_v2, VGG16, and ensemble transfer learning techniques to achieve highly accurate mask detection for individuals who are not wearing or not wearing them properly. We also propose a method to minimize training times by selecting an optimal combination of train and test dataset ratios and ensemble models for faster and real-time analysis.