Scholarlink

<p>Helmet detection is essential in engineering measurements, ensuring compliance with safety standards and providing real-time data for project management on construction sites. This paper introduces PEG-YOLOv10m, a fast and efficient detection method designed to enhance speed while preserving accuracy. The PEG module replaces the PSA attention mechanism in the backbone network, refining region-of-interest focus, optimizing feature selection, and boosting detection precision. In the neck network, redundant feature layers are removed, simplifying the model and further accelerating detection. To boost the model's effectiveness on hard samples and achieve better training outcomes, a slide loss function is used instead of binary cross-entropy loss for classification. Experimental results show that PEG-YOLOv10m achieves a mAP50 of 93.7% with 313 FPS on the SHWD dataset. Compared to YOLOv10m, PEG-YOLOv10m reduces parameters by 12%, raises mAP50 by 0.5%, and increases detection speed by 8%.</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

Fast and effective helmet detection in construction sites based on PEG-YOLOv10m

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Peilin Liu,
  • Yinfeng Fang,
  • Xuguang Zhang

摘要

Helmet detection is essential in engineering measurements, ensuring compliance with safety standards and providing real-time data for project management on construction sites. This paper introduces PEG-YOLOv10m, a fast and efficient detection method designed to enhance speed while preserving accuracy. The PEG module replaces the PSA attention mechanism in the backbone network, refining region-of-interest focus, optimizing feature selection, and boosting detection precision. In the neck network, redundant feature layers are removed, simplifying the model and further accelerating detection. To boost the model's effectiveness on hard samples and achieve better training outcomes, a slide loss function is used instead of binary cross-entropy loss for classification. Experimental results show that PEG-YOLOv10m achieves a mAP50 of 93.7% with 313 FPS on the SHWD dataset. Compared to YOLOv10m, PEG-YOLOv10m reduces parameters by 12%, raises mAP50 by 0.5%, and increases detection speed by 8%.