A novel data augmentation approach for mask detection using deep transfer learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Public Health

Background:

• The COVID-19 pandemic necessitated widespread adoption of public health measures like mask-wearing.
• Reopening economies requires technological solutions to ensure safety and compliance with health protocols.
• Effective mask detection models are vital for enforcing safety regulations in public spaces.

Purpose of the Study:

• To develop and evaluate a mask detection model using deep transfer learning.
• To enhance the performance of mask detection through a novel data augmentation technique.
• To compare the efficacy of the augmented model against a standard approach.

Main Methods:

• Utilized the YOLOv3 object detection algorithm, a deep transfer learning technique.
• Implemented a data augmentation strategy involving grayscale and Gaussian blur image filtering.
• Trained the model on both standard and augmented datasets for mask detection.
• Evaluated model performance using metrics such as mean average precision and average confidence levels.

Main Results:

• The data augmentation-based mask detection model achieved a mean average precision of 99.8% during training.
• The augmented model demonstrated higher average confidence levels across various scenarios (individuals, groups, videos) compared to the standard model.
• Specifically, confidence levels increased from 0.94 to 0.97 for individuals, 0.93 to 0.96 for groups, and 0.91 to 0.93 for videos.

Conclusions:

• The proposed data augmentation approach significantly improves the performance of YOLOv3-based mask detection.
• The enhanced model offers greater reliability in identifying mask compliance in diverse real-world settings.
• This technology serves as a valuable tool for public health enforcement and safety management.