Scholarlink

<p>This paper proposes an improved rapid discharge monitoring algorithm based on the YOLOv5 framework for real-time partial discharge detection and localization in GIS using fiber scopes and cameras. The algorithm first introduces Ghost modules to optimize the feature extraction backbone network, effectively enhancing feature extraction capability while achieving model lightweighting. Subsequently, the BiFPN structure is incorporated to improve the feature fusion process in the network neck, strengthening the model’s perceptual capacity for multi-scale features. Finally, the Meta-ACON activation function replaces the original activation functions, significantly boosting detection accuracy. Experimental results indicate that compared to the baseline YOLOv5, the proposed RDDNet achieves a detection speed of 61 frames/s with 77.2% mean average precision, meeting the requirements for real-time recognition and precise localization of internal GIS discharges.</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 discharge detection based on improved YOLOv5 algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月29日

查看全文
  • Yanhui Meng,
  • Jining Zhao,
  • Zhilong Zhao,
  • Xiaodan Fan,
  • Zhanning Wang,
  • Lei Zhao,
  • Keyu Li

摘要

This paper proposes an improved rapid discharge monitoring algorithm based on the YOLOv5 framework for real-time partial discharge detection and localization in GIS using fiber scopes and cameras. The algorithm first introduces Ghost modules to optimize the feature extraction backbone network, effectively enhancing feature extraction capability while achieving model lightweighting. Subsequently, the BiFPN structure is incorporated to improve the feature fusion process in the network neck, strengthening the model’s perceptual capacity for multi-scale features. Finally, the Meta-ACON activation function replaces the original activation functions, significantly boosting detection accuracy. Experimental results indicate that compared to the baseline YOLOv5, the proposed RDDNet achieves a detection speed of 61 frames/s with 77.2% mean average precision, meeting the requirements for real-time recognition and precise localization of internal GIS discharges.