Facial Mask Detection Using Depthwise Separable Convolutional Neural Network Model During COVID-19 Pandemic.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Deep neural networks (DNNs) excel at facial photo categorization, but challenges remain due to feature complexity, large image sizes, and data inhomogeneity.
• Efficient face image classification is crucial, especially in mobile contexts, with growing data volumes and the need for advanced deep learning (DL) techniques.
• Existing DL approaches for face identification often employ Convolutional Neural Networks (CNNs).

Purpose of the Study:

• To propose an efficient face mask recognition method using a novel Depthwise Separable Convolution Neural Network (DWS-based MobileNet).
• To address the limitations of traditional CNNs in handling complex facial image data for classification tasks.
• To develop a lightweight network that enhances learning performance and reduces trainable parameters for effective face image categorization.

Main Methods:

• Implementation of a DWS-based MobileNet architecture utilizing depth-wise separable convolution layers instead of standard 2D convolution layers.
• Development of a lightweight network designed to decrease the number of trainable parameters while maintaining high learning performance.
• Evaluation of the proposed DWS-based MobileNet on benchmark datasets for face mask recognition.

Main Results:

• The DWS-based MobileNet demonstrated exceptional performance, particularly with limited datasets.
• The proposed network significantly reduced trainable parameters compared to Full Convolution MobileNet and baseline methods.
• Achieved high performance metrics: Accuracy = 93.14%, Precision = 92%, Recall = 92%, and F-score = 92%.

Conclusions:

• The DWS-based MobileNet offers a significant improvement over existing state-of-the-art methods for face mask recognition.
• The lightweight architecture and use of depth-wise separable convolutions make it highly effective for efficient face image classification.
• This approach is well-suited for mobile applications requiring robust and accurate facial recognition, even with masked individuals.