Abstract

Detecting face masks is essential for maintaining public safety and preventing the spread of contagious illnesses. In this article, we give a thorough inves- tigation into how Convolutional Neural Networks (CNNs) may improve face mask identification. The goal of this work is to provide a reliable and robust CNN-based method for identifying people who are wearing masks in practical situations. We start by outlining the CNN architecture, which has a sequen- tial structure made up of convolutional layers, activation functions, pooling layers, and fully linked layers, and is utilized for facemask identification. The architecture is made to recognize masked and unmasked faces with accuracy and learn hierarchical representations of input photos. Layers are pooled for downsampling, fully linked layers are used for high-level representations, and activation functions are used to induce non-linearities. We use a number of measures, including accuracy, precision, recall, and F1-score, to assess the performance of our CNN model. The accuracy of our experimental findings is encouraging, with a 95% overall accuracy in identifying people wearing masks. The accuracy in accurately detecting both positive and negative cases is balanced, as seen by the precision and recall values, which are determined to be 92% and 96%, respectively.  We also assess the model’s effectiveness in other scenarios, such as those involving several people spread out across a wide area. Our findings show that even when people are at different distances from one another, there is constant performance with a high accuracy rate of above 90%. This demonstrates the model’s capacity to identify masks regardless of the distance that people are from the camera. We compare the performance of our CNN-based approach to current mask recognition algorithms and show how it outperforms them, outperforming more conventional approaches that generally had accuracy levels of 70–80%.