Multi-Scene Mask Detection Based on Multi-Scale Residual and Complementary Attention Mechanism

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• Optical sensors generate vast data, enabling deep learning for real-world applications like mask detection.
• Challenges in mask detection include small targets, complex scenes, and occlusions, requiring robust multi-scene detection networks.

Purpose of the Study:

• To propose an accurate and robust deep learning network for multi-scene mask detection.
• To enhance detection of small targets and improve performance in complex environments.

Main Methods:

• Developed the YOLO-MSM network using multi-scale residual (MSR) blocks and attention mechanisms (MSR-CCSA, ER-CCSA, ER-PCSA).
• Utilized YOLOv5 as the baseline and introduced hierarchical residual connections for multi-scale feature extraction.
• Created a new Multi-Scene-Mask dataset with diverse scenarios.
• Implemented a generalization improvement strategy (GIS) using data augmentation.

Main Results:

• YOLO-MSM achieved an average precision of 97.51% on the Multi-Scene-Mask dataset.
• Demonstrated a 3.46% increase in mean average precision (mAP) compared to the baseline YOLOv5 network.
• The GIS strategy significantly enhanced the network's generalization ability.

Conclusions:

• The proposed YOLO-MSM network offers superior performance for multi-scene mask detection.
• YOLO-MSM exhibits improved accuracy, robustness, and generalization compared to existing methods.
• The developed dataset and GIS strategy contribute to advancing mask detection research.