基于改进ED-YOLOv5s的矿井安全帽佩戴检测算法

Detection algorithm for wearing safety helmet undermine based on improved ED-YOLOv5s

煤矿井下工作中安全帽佩戴是事关工人生命安全的一大关键要素。基于视频图像进行分析的技术虽可以较好地检测工人安全帽佩戴情况从而将事故带来的损害最小化,但是在矿井下的图像收集过程中往往存在各种各样的现实因素,例如环境复杂、存在多个目标等,给技术人员造成很大的干扰。针对以上问题,文章通过引进EMA注意力机制与DIoU损失函数,提出了一种改进的ED-YOLOv5s模型。在自制数据集上对该模型进行了消融实验,结果表明该模型相比原模型在图像检测速度和精度方面都有较大的提升。随后,文章将该算法与YOLOv7-tiny、YOLOv8进行对比实验,结果显示文章算法在矿井下安全帽检测的mAP@50%达到了97.3%。

Wearing a helmet in underground coal mine is a key factor concerning the safety of workers.Although the video image analysis technology can better detect the helmet wearing of workers to minimize the damage caused by accidents,there are often various realistic factors in the image collection process under the mine,such as complex environment and multiple targets.These problems will cause a lot of interference to the technicians.To address the above problems,this study proposed an improved ED-YOLOv5s model by introducing the EMA attention mechanism with the DIoU loss function.In this paper,we conducted ablation experiments on CUMT-HelmeT dataset,and the results show that it is greatly improved in image detection speed and accuracy compared with the original model.After comparing the algorithm with YOLOv7-tiny and YOLOv8,result display that the mAP@50%is 97.3%.