【摘要】 通常以人工查验方式来检验公共场所进出人员是否规范佩戴口罩，计算机视觉技术辅助的自动查验大多用于佩戴口罩时的人脸识别、是否佩戴口罩等特定场景人物，而缺乏对口罩佩戴规范性的自动检测。针对口罩佩戴规范性的视觉检测问题，提出了一种基于改进YOLOv5s框架的口罩佩戴规范性检测模型。为提高口罩佩戴规范性检测算法的准确率和稳定性，在paddle口罩数据集基础上，通过互联网检索，并经过人工筛选与标注后，构建了口罩佩戴规范性检测数据集，提出了基于改进YOLOv5s的口罩佩戴规范性视觉检测模型。该模型融合了多种嵌入注意力机制，实验结果表明，提出的YOLOv5s-ECA模型的先进性可以满足实际场景的应用需求。更多还原

【Abstract】 Manual inspection is usually used to verify whether personnel entering and exiting public places wear masks in a standardized manner. Computer vision assisted automatic inspection is mostly used for face recognition when wearing masks, whether to wear masks and other specific scenes,and there is a lack of standardized automatic detection of wearing masks. A mask wearing norm detection model based on improved YOLOv5s framework was proposed to address the issue of mask wearing norm. In order to improve the accuracy and stability of mask wearing standardization detection algorithm, based on the paddle mask data set, the mask wearing standardization detection data set was constructed through internet retrieval, manual screening and annotation, and a mask wearing normative visual detection model based on improved YOLOv5s was proposed. The model integrates a variety of embedded attention mechanisms, and the experimental results show that the advanced nature of the proposed YOLOv5s-ECA model can meet the application requirements of actual scenarios.更多还原