SCS-Net: An efficient and practical approach towards Face Mask Detection.

Area of Science:

• Computer Vision
• Deep Learning
• Public Health

Background:

• Facial mask detection is crucial for curbing Coronavirus disease (COVID-19) spread.
• Deep learning models show high accuracy but lack real-world deployability and struggle with incorrectly worn masks.

Purpose of the Study:

• To develop a lightweight deep learning model for practical facial mask detection.
• To address the challenge of incorrectly worn masks by augmenting datasets.
• To create a more deployable and accurate solution for public health.

Main Methods:

• Proposed a novel, lightweight deep learning architecture with 0.12M parameters.
• Augmented an existing dataset with 25,296 synthetically generated images of incorrectly worn masks.
• Developed a three-class classification system for mask detection.

Main Results:

• Achieved 95.41% accuracy on two-class classification and 95.54% on three-class classification.
• The model has a parameter reduction of up to 496 times compared to existing models.
• Demonstrated superior real-world applicability and rationality over massive, less deployable models.

Conclusions:

• The developed lightweight model offers a practical and efficient solution for facial mask detection.
• The augmented dataset and three-class classification improve robustness to incorrectly worn masks.
• This approach enhances the deployability of deep learning for public health interventions.