Research of mask wearing detection method using Improved YOLO network

Mask wearing detection technology is a practical and research-oriented area within object detection. Due to the complexity of mask wearing detection scenarios, issues such as small target detection and low image clarity often result in missed detections and false alarms. The YOLO algorithm, known for its fast detection speed, is commonly used in mask wearing detection. However, the original YOLO algorithm struggles to achieve satisfactory results in dense crowd scenarios. Therefore, this paper proposes an improved YOLOv5 algorithm based on the BiFPN structure and incorporating attention mechanisms to enhance the accuracy of mask wearing detection while maintaining real-time performance. The proposed approach firstly introduces the Bidirectional Feature Pyramid (BiFPN) structure into the backbone network to enhance the feature extraction capability. Additionally, attention mechanisms are incorporated to filter out irrelevant information, conserve computational resources, and improve the detection performance for small targets. Experimental results on a custom dataset demonstrate that the YOLOv5_BiFPN algorithm achieves a 4% improvement in accuracy and an 11.7% increase in mAP50 compared to the original YOLOv5. Moreover, the inclusion of attention mechanisms further boosts the accuracy by an additional 1.6%. These findings confirm that the improved YOLOv5s model exhibits superior performance in the task of mask wearing detection.