Abstract
Face mask wearing detection is an important application scenario in current technology. This study proposes a method based on the YOLOv5 object detection algorithm to address this issue. Traditional methods face challenges such as the diversity of mask-wearing postures and variations in lighting conditions, which affect their performance. To tackle these challenges, this research presents a new approach that combines the YOLOv5 object detection algorithm with an improved ResNet network architecture. By integrating the detection capabilities of YOLOv5 with the enhanced ResNet network, the method can more accurately detect masks and their wearing status, effectively capturing mask features in images, thereby significantly improving recognition accuracy and stability. The use of a custom mask dataset enables the model to better adapt to diverse lighting and posture conditions. Using deep learning frameworks like PyTorch for inference tools has significantly improved inference speed on GPUs. Experimental results show that after 200 training epochs, the proposed method achieved an accuracy exceeding 85% in face mask wearing detection tasks, with detection accuracy surpassing 98% on certain test datasets. Furthermore, the mean average precision (mAP) reached 97.5%, demonstrating the model's robustness under complex backgrounds and diverse populations. Finally, this paper discusses potential future development directions in the field of face mask wearing detection, including further enhancing the model's adaptability to varying environmental conditions and its application in real-time detection systems.