Scholarlink

<p>Timely detection of transformer bushing faults is crucial for the safe operation of power systems. This article introduces LiteYOLO-GHG, a lightweight fault detection model based on YOLOv8s. The original YOLOv8 backbone is replaced with GhostHGNetV2, which enhances detection accuracy, while significantly reducing model parameters. A lightweight HGStem module is employed to improve feature extraction and further decrease parameters. Additionally, the original YOLOv8s detection head is substituted with a lightweight head, D-Eff, to boost accuracy without substantial increases in parameters. Experimental result demonstrate that LiteYOLO-GHG achieved an mAP@0.5 of 90.8%, reflecting 3.06% improvement in accuracy, alongside reductions in parameters, computational complexity, and model size by 21.39, 32.39, and 18.18%, respectively. These findings underscore the effectiveness and accuracy of LiteYOLO GHG as a lightweight model for transformer bushing fault detection algorithm.</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

LiteYOLO-GHG: a lightweight YOLOv8-based algorithm for transformer bushing fault detection

  • 期刊论文

  • 发表日期：

    2025年1月4日

查看全文
  • Senyue Xiao,
  • Jianhua Liu,
  • Zeming Pan,
  • Shaoze Wang,
  • Yang Yang,
  • Zilong Song,
  • Anni Fan

摘要

Timely detection of transformer bushing faults is crucial for the safe operation of power systems. This article introduces LiteYOLO-GHG, a lightweight fault detection model based on YOLOv8s. The original YOLOv8 backbone is replaced with GhostHGNetV2, which enhances detection accuracy, while significantly reducing model parameters. A lightweight HGStem module is employed to improve feature extraction and further decrease parameters. Additionally, the original YOLOv8s detection head is substituted with a lightweight head, D-Eff, to boost accuracy without substantial increases in parameters. Experimental result demonstrate that LiteYOLO-GHG achieved an mAP@0.5 of 90.8%, reflecting 3.06% improvement in accuracy, alongside reductions in parameters, computational complexity, and model size by 21.39, 32.39, and 18.18%, respectively. These findings underscore the effectiveness and accuracy of LiteYOLO GHG as a lightweight model for transformer bushing fault detection algorithm.