Hybrid Deep Feature Generation for Appropriate Face Mask Use Detection.

Area of Science:

• Computer Science
• Public Health

Background:

• Mask usage is critical for limiting COVID-19 transmission.
• Ensuring correct face covering use is essential for public health compliance.
• Automated systems can enhance monitoring of mask-wearing behaviors.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for classifying appropriate face mask usage.
• To address the challenge of inappropriate mask use through automated detection.
• To assess the model's performance across different classification scenarios.

Main Methods:

• Collected and labeled 2075 images of face mask usage (mask, no mask, improper mask).
• Employed a hybrid deep feature-based model using pre-trained ResNet101 and DenseNet201 as feature generators.
• Utilized an improved RelieF selector for feature selection and a support vector machine classifier.

Main Results:

• Achieved classification accuracy rates of 95.95% (Case 1: mask vs. no mask vs. improper mask), 97.49% (Case 2: mask vs. no mask + improper mask), and 100.0% (Case 3: mask vs. no mask).
• Demonstrated high performance in distinguishing between appropriate and inappropriate mask usage.
• The model's accuracy indicates suitability for real-time applications.

Conclusions:

• The proposed hybrid deep feature-based model effectively classifies face mask usage with high accuracy.
• This technology shows promise for practical implementation in real-time compliance monitoring systems.
• Automated classification can significantly contribute to enforcing public health guidelines during pandemics.