基于改进YOLOv5目标检测算法的口罩识别

Mask Recognition Based on an Improved YOLOv5 Target Detection Algorithm

为遏制疫情蔓延,正确佩戴口罩是有效的防护手段之一,但人工检测和监督的效率很低。基于深度学习网络模型的口罩实时自动检测在准确性和实时性方面还有待提升。在YOLOv5目标检测算法的基础上,创建人脸识别数据集MaskData。采用DIOU_nms方法替代NMS中的IOU,提高对遮挡和重叠目标的检测精度。最后,用α-CIoU损失函数替换GIOU损失函数,更快、更准确地获得更高质量的定位图像区域,生成边界框并预测类别。改进后的YOLOv5检测网络,对创建的数据集MaskData进行训练。实验结果表明,改进后的网络在94.45%的检测精度的基础上,提升到了95.2%,可以更准确地识别人脸是否戴口罩。

To curb the spread of the epidemic,proper wearing of masks is one of the most effective means of protection,but manual detection and supervision are inefficient.There is still room for improvement in accuracy and in real-time automatic detection of masks based on deep learning network models.In this paper,MaskData,a face recognition dataset,is created based on the YOLOv5 target detection algorithm.The DIOU_nms method is used to replace IOU in NMS in the detection network to improve the detection accuracy of masked and overlapping targets.The GIOU loss function is replaced with theα-CIoU loss function to obtain higher-quality localized image regions,generate bounding boxes and predict categories faster and more accurately.The improved YOLOv5 detection network is trained on the created dataset MaskData.The experimental results show that the improved network can also improve detection accuracy from 94.45%to 95.2%,it can more accurately identify whether a face is wearing a mask or not.