Scholarlink

Electric power is the significant support of national economy. However, large steel structures of the power stations in coastal areas are facing to harsh atmospheric corrosion conditions in long term. It may decrease the reliability of devices and brings high safety risks. As one commonly used maintenance method, manual inspection is time-consuming, labor-intensive, and poses severe personal safety risks. The advancement of computer vision techniques offers a rapid and accurate non-contact alternative for detecting corrosion on such structures. In this paper, we addressed the limitations of existing deep learning methods for accurate detection of corrosion from the interference such as aged coating yellowing by proposing an improved YOLOv8n model, YOLOv8-RM. The method enhanced the extraction of corrosion features in complex backgrounds by incorporating the RepNCSPELAN4 module and the MC attention mechanism module, leading to improved detection accuracy. Comparative experiments demonstrate that the proposed YOLOv8-RM algorithm outperforms other methods, achieving a detection accuracy of 91.8%, while also meeting the speed requirements for deployment on edge devices.

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

High-Efficient Automatic Corrosion Detection for Large Steel Structures of Power Stations in Coastal Area Based on an Improved YOLOv8 Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jun Wang,
  • Jiaxu Duan,
  • Miaoran Liu,
  • Xu Zhang,
  • Yunke Nie,
  • Xin Wu

摘要

Electric power is the significant support of national economy. However, large steel structures of the power stations in coastal areas are facing to harsh atmospheric corrosion conditions in long term. It may decrease the reliability of devices and brings high safety risks. As one commonly used maintenance method, manual inspection is time-consuming, labor-intensive, and poses severe personal safety risks. The advancement of computer vision techniques offers a rapid and accurate non-contact alternative for detecting corrosion on such structures. In this paper, we addressed the limitations of existing deep learning methods for accurate detection of corrosion from the interference such as aged coating yellowing by proposing an improved YOLOv8n model, YOLOv8-RM. The method enhanced the extraction of corrosion features in complex backgrounds by incorporating the RepNCSPELAN4 module and the MC attention mechanism module, leading to improved detection accuracy. Comparative experiments demonstrate that the proposed YOLOv8-RM algorithm outperforms other methods, achieving a detection accuracy of 91.8%, while also meeting the speed requirements for deployment on edge devices.