Advanced Infrared Face Mask Segmentation Using a Custom Lightweight U-Net Model

Face mask segmentation in the infrared domain rep-resents an innovative technique aimed at improving the accuracy and efficiency of detecting and isolating face masks in thermal images. This approach holds significant relevance in various applications, such as enhancing public health by verifying mask usage in medical environments, improving security through more accurate facial recognition of individuals wearing masks, and advancing human-computer interaction by enabling hands-free device control in mask-mandatory settings, thereby promoting safety and hygiene. In this research, we propose a novel solution that employs deep learning algorithms with infrared imaging to overcome the limitations of conventional mask detection methods. The core of this solution is a custom-designed, optimized variant of the U-Net model. Furthermore, this study forms part of a larger project focused on developing stations for detecting and quantifying leaks in medical masks across various mask types using infrared technologies and artificial intelligence.