Face Mask Classification Based on the Convolutional Neural Network

Abstract

The worldwide emergence of the COVID-19 pandemic has instigated widespread anxiety. In an effort to curb the virus's transmission, the adoption of mask-wearing has steadily become more prevalent. Numerous public facilities now require individuals to wear masks upon entry and exit. Within this framework, the effective and economical identification of mask-wearing individuals has become a growing priority, and deep learning has demonstrated its capability to address this need. Utilizing a dataset consisting of 440 provided images, the current experiment employed a Convolutional Neural Network (CNN) model to determine the presence or absence of masks on individuals. The experimental process encompassed dataset segmentation, CNN model construction, model training, and performance evaluation with the objective of attaining the best-performing model. The research revealed that, within a specific range, an increase in the number of layers corresponded to an enhancement in the CNN's performance. Remarkably, at 13 layers, the CNN achieved its highest accuracy, reaching 93.18%, with a recorded lowest loss value of 0.1608. Despite this, the CNN model's accuracy decreased as a result of an increase in the convolutional kernel size, which was accompanied by variations in the loss value. After multiple iterations of experimentation, the optimal convolutional kernel size was determined to be 3x3. In the quest for a more effective means of detecting mask compliance in public spaces, this study underscores the potential of deep learning techniques, particularly CNNs, and highlights the significance of carefully configuring model architecture for optimal results.