FMD-Yolo: An efficient face mask detection method for COVID-19 prevention and control in public

Area of Science:

• Computer Vision
• Artificial Intelligence
• Public Health

Background:

• The COVID-19 pandemic necessitates effective public health interventions, including widespread face mask usage.
• Monitoring correct mask-wearing in public is vital for controlling virus transmission.

Purpose of the Study:

• To introduce FMD-Yolo, a novel framework for accurate and efficient face mask detection.
• To enhance the robustness and generalization ability of face mask detection models.

Main Methods:

• Utilized Im-Res2Net-101 as a feature extractor, incorporating Res2Net, deep residual networks, and non-local mechanisms.
• Employed an enhanced path aggregation network (En-PAN) for effective feature fusion.
• Implemented a designed localization loss and Matrix Non-Maximum Suppression (NMS) for improved detection.

Main Results:

• FMD-Yolo achieved superior performance on two benchmark datasets compared to eight state-of-the-art algorithms.
• Achieved the highest precision (AP50) of 92.0% and 88.4% at IoU=0.5.
• Demonstrated significant improvements in AP75 at IoU=0.75, with increases of 5.5% and 3.9%.

Conclusions:

• FMD-Yolo exhibits significant superiority in face mask detection, offering both theoretical and practical value.
• The proposed framework effectively addresses the need for reliable public mask-wearing monitoring.
• This technology holds practical significance for public health surveillance and disease prevention efforts.