Scholarlink

<p>The stability of transmission lines is essential for reliable power delivery, yet abnormal targets such as bird nests and foreign objects pose serious operational hazards. While UAV-based inspections combined with deep learning models, particularly the YOLO series, have advanced anomaly detection, existing approaches often struggle to simultaneously achieve high accuracy, real-time performance, and efficient deployment on edge devices. To overcome these limitations, we present YOLO-TLD, a lightweight and high-precision detection model built upon YOLOv8. YOLO-TLD incorporates three key innovations: an Efficient Channel Attention (ECA) module to strengthen feature representation and suppress background noise; a CGB_Down module to fuse local and global contextual features; and an Inner-CIoU loss function to enhance bounding box regression accuracy through auxiliary boundary constraints. These improvements collectively increase robustness in complex scenes while preserving low computational overhead. Extensive experiments on a custom-built dataset and the public RailFOD23 benchmark demonstrate that YOLO-TLD consistently surpasses its baseline in detection performance while maintaining a compact architecture and real-time inference capability. Overall, YOLO-TLD offers a compelling solution for UAV-based transmission line anomaly detection, balancing accuracy, efficiency, and deployment feasibility.</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

YOLO-TLD: transmission line abnormal target detection model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月19日

查看全文
  • Long Ling,
  • Jingde Huang,
  • Yumeng Lu

摘要

The stability of transmission lines is essential for reliable power delivery, yet abnormal targets such as bird nests and foreign objects pose serious operational hazards. While UAV-based inspections combined with deep learning models, particularly the YOLO series, have advanced anomaly detection, existing approaches often struggle to simultaneously achieve high accuracy, real-time performance, and efficient deployment on edge devices. To overcome these limitations, we present YOLO-TLD, a lightweight and high-precision detection model built upon YOLOv8. YOLO-TLD incorporates three key innovations: an Efficient Channel Attention (ECA) module to strengthen feature representation and suppress background noise; a CGB_Down module to fuse local and global contextual features; and an Inner-CIoU loss function to enhance bounding box regression accuracy through auxiliary boundary constraints. These improvements collectively increase robustness in complex scenes while preserving low computational overhead. Extensive experiments on a custom-built dataset and the public RailFOD23 benchmark demonstrate that YOLO-TLD consistently surpasses its baseline in detection performance while maintaining a compact architecture and real-time inference capability. Overall, YOLO-TLD offers a compelling solution for UAV-based transmission line anomaly detection, balancing accuracy, efficiency, and deployment feasibility.