基于改进YOLOv5s的口罩佩戴检测算法

Mask wearing detection algorithm based on improved YOLOv5s

在新冠疫情的影响下,佩戴口罩成为人们日常必备的防护措施。为了更好地实现智能化管理,针对公共场合密集人群佩戴口罩是否正确检测任务中的过小目标检测和遮挡问题,提出了一种基于改进YOLOv5s的实时检测算法,通过引入自注意力机制,从而提高模型的显著特征,进而优化算法精度;改变Neck层的卷积结构,采用基于双尺度的特征融合目标检测技术,实现了更好地特征提取。通过对改进后的YOLOv5s算法进行试验,证明了该方法模型小、检测速度快,并且平均识别精度均值比原来的方法提高了4.4%,更好地解决了复杂背景下、目标检测任务中过小目标的检测和遮挡问题。

Under the influence of Covid-19,wearing masks has become a daily necessary protection measure for people.In order to better realize the intelligent management,in view of the public crowd wearing masks is correct detection of small target detection and occlusion problem,this paper proposes a real-time detection algorithm based on improved YOLOv5 s,by introducing the concentration mechanism,so as to improve the characteristic of the model,and optimized algorithm accuracy;The convolution structure of neck layer is changed,and the feature fusion object detection based on double scale is adopted to achieve better feature extraction.Through experiments on the improved YOLOv5 s algorithm,it is proved that the proposed method has small model,fast detection speed and the average recognition accuracy is 4.4%higher than before,which can better solve the problem of detection and occlusion of too small targets in complex background.