Scholarlink

<p>In heavy machinery factories, accurately detecting whether workers correctly wear safety helmets is important to their well-being. Since manual inspection and video surveillance are prone to misjudgment and omission, designing a fast and intelligent algorithm essential for modern factory safety management. The YOLO series, a popular object location and detection method, offers an excellent balance between detection speed and accuracy, drawing wide attention from industry scholars. In light of this, this paper presents an improved model based on YOLOv10 to achieve safety helmet identification. Firstly, it replaces Conv convolution with distributed shift DSConv convolution in YOLOv10. This boosts memory efficiency in the convolutional layer and ensures small object identification accuracy. Secondly, the Dysample module is incorporated to cut computational load, enhance sampling, and improve model generalizability. Additionally, the WIoU loss function is introduced to accelerate convergence and increase adaptability. When compared with mainstream object recognition algorithms such as SSD, Faster RCNN, and various YOLO versions, the optimized model shows its superiority. Compared to the original YOLOv10, its average accuracy rises by 0.5%, while floating-point computation and model size decrease by 7.1% and 1.4% respectively. Finally, the optimized model is deployed on the Atlas200I DK A2 computing box to validate its usability on IoT edge devices.</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

Safety helmet detection methods in heavy machinery factory

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月27日

查看全文
  • Liu Baoju,
  • Wei Xiangqian,
  • Chen Qingshan,
  • Liu Jiaqi,
  • Chen Ye,
  • Yu Peng,
  • Lei shi,
  • Hu Yongfeng

摘要

In heavy machinery factories, accurately detecting whether workers correctly wear safety helmets is important to their well-being. Since manual inspection and video surveillance are prone to misjudgment and omission, designing a fast and intelligent algorithm essential for modern factory safety management. The YOLO series, a popular object location and detection method, offers an excellent balance between detection speed and accuracy, drawing wide attention from industry scholars. In light of this, this paper presents an improved model based on YOLOv10 to achieve safety helmet identification. Firstly, it replaces Conv convolution with distributed shift DSConv convolution in YOLOv10. This boosts memory efficiency in the convolutional layer and ensures small object identification accuracy. Secondly, the Dysample module is incorporated to cut computational load, enhance sampling, and improve model generalizability. Additionally, the WIoU loss function is introduced to accelerate convergence and increase adaptability. When compared with mainstream object recognition algorithms such as SSD, Faster RCNN, and various YOLO versions, the optimized model shows its superiority. Compared to the original YOLOv10, its average accuracy rises by 0.5%, while floating-point computation and model size decrease by 7.1% and 1.4% respectively. Finally, the optimized model is deployed on the Atlas200I DK A2 computing box to validate its usability on IoT edge devices.