摘要
针对矿井作业中人员防护装备检测准确度不足、响应速度缓慢等问题,本研究提出了一种改进YOLOv8n的矿井人员防护装备检测算法,命名为EF-YOLOv8n。EF-YOLOv8n模型首先将基准模型主干网络C2f中的瓶颈结构模块用Faster模块替换,得到了C2f-Faster模块,有利于提高模型的检测精度,降低模型的体积,提高模型的运算速度。其次在最后一个C2f-Faster模块后引入了EMA模块。可以有效提高模型对于小目标的检测能力。与YOLOv8n模型相比EF-YOLOv8n模型mAP@50提升2.4%,mAP@50~95提升1.6%,参数量减少0.4%,模型体积下降0.7%。本模型降低了漏检的发生几率,提高了模型的检测能力,为守护矿工们的安全提供了坚实的科技保障。
In response to the issues of insufficient accuracy and slow response speed in the detection of personal protective equipment(PPE)during mine operations,this study proposes an improved mine PPE detection algorithm based on YOLOv8n,named EF-YOLOv8n.The EF-YOLOv8n model first replaces the Bottleneck module in the backbone network C2f of the baseline model with a Faster Net Block module,resulting in the C2f-Faster module.This modification enhances the detection accuracy of the model,reduces its size,and increases computational speed.Secondly,an EMA module is introduced after the final C2f-Faster module,effectively improving the model's ability to detect small targets.Compared to the YOLOv8n model,the EF-YOLOv8n model achieves a 2.4%increase in mAP@50,a 1.6%increase in mAP@50~95,a 0.4%reduction in parameters,and a 0.7%decrease in model size.This model reduces the likelihood of missed detections and enhances the detection capability,providing solid technological support for safeguarding the safety of miners.
作者
孙晓岩
张磊
叶军建
孙志鹏
郑奥
李平
SUN Xiaoyan;ZHANG Lei;YE Junjian;SUN Zhipeng;ZHENG Ao;LI Ping(School of Coal Engineering,Shanxi Datong University,Datong Shanxi,037000)
出处
《山西大同大学学报(自然科学版)》
2025年第4期28-33,共6页
Journal of Shanxi Datong University(Natural Science Edition)
关键词
EMA
目标检测
深度学习
防护装备
EMA
target detection
deep learning
protective equipment
