基于YOLOv5s的口罩佩戴实时检测系统设计  

Design of Mask Wearing Real-time Detection System Based on YOLOv5s

作　　者：焦双健[1] 谢似霞 JIAO Shuangjian;XIE Sixia(Faculty of Engineering,Ocean University of China,Qingdao 266100,China)

机构地区：[1]中国海洋大学工程学院,山东青岛266100

出　　处：《电视技术》2022年第11期52-56,60,共6页Video Engineering

摘　　要：正确佩戴防护口罩,已经成为一种科学有效的降低病毒传播率的方法。然而,现有的人工检查是否正确佩戴口罩的方法存在明显不足。基于目前目标检测技术的快速发展,为了节省人力、避免漏检和提高防控效率,提出一种基于YOLOv5s的口罩佩戴实时检测系统,用于实时检测人们是否佩戴口罩。改进的方法加入了SE注意力机制,改进了目标框回归损失函数。该系统将训练好的YOLOv5s模型部署到相关设备上以实现实际应用,检测速度快,检测准确率高。当检测到未佩戴口罩的人脸时,系统会马上发出警报,对未佩戴口罩的人提供快速的提醒。Wearing the respirator correctly has become a scientific and effective method to reduce the virus transmission rate. However,the existing manual inspection methods for correct wearing of masks have obvious shortcomings. Based on the rapid development of target detection technology, in order to save manpower, avoid missing detection and improve the efficiency of prevention and control, a real-time detection system for mask wearing based on YOLOv5s is proposed, which is used to detect whether people wear masks in real time. The improved method adds SE attention mechanism and improves the target box regression loss function. The system deploys the trained YOLOv5s model to relevant equipment for practical application, with fast detection speed and high detection accuracy. When a face without a mask is detected, the system will send an alarm immediately to provide a quick reminder to the person without a mask.

关 键 词：口罩检测 目标检测 YOLOv5s 注意力机制 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]