基于改进YOLOv5的人脸口罩佩戴检测

Face mask wearing detection based on improved YOLOv5 algorithm

为对商场、车站等复杂环境中的人脸口罩佩戴情况进行检测,综合考虑目标密集、遮挡和小尺度目标等因素,提出一种复杂环境下基于改进YOLOv5的人脸口罩检测方法。引入改进DenseNet(密集连接卷积网络),提高网络特征利用率以及网络抗干扰能力;增加检测头部参数,对不同尺度特征跨级连接,增强多尺度信息交流,提高网络对小尺度目标的检测性能;将原有损失函数GIoU替换为CIoU,解决模型收敛速度慢的问题。实验结果表明,在人脸口罩佩戴检测任务中,改进YOLOv5算法mAP(平均精度均值)为97.8%,较YOLOv5算法与其它主流算法具有更高的检测精度,对实际场景中的人脸口罩检测任务具有现实意义。

To detect the wearing of face masks in shopping malls,stations and other environments,a face mask detection method based on improved YOLOv5 was proposed in complex environments.The factors such as dense targets,occlusion and small scale were considered.The improved DenseNet network was introduced,network feature utilization and network anti-interference ability were improved.The number of head detection parameters was increased,and different scale features across levels were connected,multi-scale information exchange was enhanced,and the network’s detection performance for small-scale targets was improved.The original loss function GIoU was replace with CIoU,which solved the problem of slow model convergence.Experi-mental results show that,in the face mask wearing detection task,the mAP(average mean precision)of the improved YOLOv5 algorithm is 97.8%,which shows higher detection accuracy than the YOLOv5 algorithm and other mainstream algorithms.The proposed method has practical significance.