Scholarlink

<p>Tunnel construction environments present a confluence of hazardous conditions, including poor lighting, high levels of dust, and frequent occlusion from machinery, which significantly increase the risk of accidents. Ensuring workers correctly wear personal protective equipment (PPE) is a critical safety measure, yet traditional manual supervision is inefficient and prone to oversight. To address this challenge, this paper introduces an improved YOLOv5-based method specifically designed for robust PPE detection in low-light tunnel environments. Our approach integrates three key innovations: a channel-metric (CM) attention mechanism to enhance feature contrast in dark conditions; an adaptive feature pyramid network (AFPN) to improve the detection of small and occluded targets; and an XIoU_NMS function to reduce missed detections in cluttered scenes. Experimental results on a real-world tunnel dataset demonstrate significant improvements, achieving a precision of 94.6% and a mean average precision (mAP@0.5) of 90.2%. The model demonstrates stable performance in actual tunnel monitoring systems, showing potential for enhancing construction safety management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced PPE detection in low-light tunnel environments: a YOLOv5-based approach

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Menglan Wu,
  • Weiwei Xie,
  • Guanxian Song,
  • Bo Yu,
  • Chuanyi Huang,
  • Yuge Lu,
  • Hao Peng,
  • Nenghao Huang,
  • Menglong Zhu,
  • Ming Liang

摘要

Tunnel construction environments present a confluence of hazardous conditions, including poor lighting, high levels of dust, and frequent occlusion from machinery, which significantly increase the risk of accidents. Ensuring workers correctly wear personal protective equipment (PPE) is a critical safety measure, yet traditional manual supervision is inefficient and prone to oversight. To address this challenge, this paper introduces an improved YOLOv5-based method specifically designed for robust PPE detection in low-light tunnel environments. Our approach integrates three key innovations: a channel-metric (CM) attention mechanism to enhance feature contrast in dark conditions; an adaptive feature pyramid network (AFPN) to improve the detection of small and occluded targets; and an XIoU_NMS function to reduce missed detections in cluttered scenes. Experimental results on a real-world tunnel dataset demonstrate significant improvements, achieving a precision of 94.6% and a mean average precision (mAP@0.5) of 90.2%. The model demonstrates stable performance in actual tunnel monitoring systems, showing potential for enhancing construction safety management.