Scholarlink

<p>This paper proposes SCCFM-YOLO, an ultra-lightweight and efficient real-time object detection framework designed for transmission line defect detection in complex environments. Built upon the YOLO11n architecture, SCCFM-YOLO integrates a ShuffleNetV2 backbone, a CCFM-SCDown neck module, a Context Anchor Attention (CAA) mechanism, the Focaler-Inner-IoU loss function, and Mask Generative Distillation (MGD) to optimize both detection accuracy and computational efficiency. Evaluated on a dedicated dataset covering three typical defect categories—cable damage, insulator loss, and tower corrosion—the proposed model achieves a mean average precision (mAP50) of 92%, with only 0.914&#xa0;M parameters and a real-time speed of 207 FPS, significantly outperforming state-of-the-art lightweight detection models such as YOLOv8 and YOLOv10. The detection speed of the SCCFM-YOLO model has been randomly tested on 200 images, and the detection time for each image is only 3.67 ms, which meets the real-time and accurate detection standard. The robustness of the SCCFM-YOLO model under different lighting conditions is demonstrated through five lighting experiments conducted on the same photograph. The model also demonstrates a significant improvement in accuracy under complex backgrounds. On the CPLID public dataset, the model increased the mAP50 by 9.6% compared to the baseline model.</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

Real-time detection method of typical defects in transmission line under complex lighting and backgrounds

  • 期刊论文

  • 发表日期：

    2025年10月17日

查看全文
  • Guoguang Tan,
  • Liu Liu,
  • Ziteng Liu,
  • Bin Wen,
  • Lili Li,
  • Yongsheng Ye

摘要

This paper proposes SCCFM-YOLO, an ultra-lightweight and efficient real-time object detection framework designed for transmission line defect detection in complex environments. Built upon the YOLO11n architecture, SCCFM-YOLO integrates a ShuffleNetV2 backbone, a CCFM-SCDown neck module, a Context Anchor Attention (CAA) mechanism, the Focaler-Inner-IoU loss function, and Mask Generative Distillation (MGD) to optimize both detection accuracy and computational efficiency. Evaluated on a dedicated dataset covering three typical defect categories—cable damage, insulator loss, and tower corrosion—the proposed model achieves a mean average precision (mAP50) of 92%, with only 0.914 M parameters and a real-time speed of 207 FPS, significantly outperforming state-of-the-art lightweight detection models such as YOLOv8 and YOLOv10. The detection speed of the SCCFM-YOLO model has been randomly tested on 200 images, and the detection time for each image is only 3.67 ms, which meets the real-time and accurate detection standard. The robustness of the SCCFM-YOLO model under different lighting conditions is demonstrated through five lighting experiments conducted on the same photograph. The model also demonstrates a significant improvement in accuracy under complex backgrounds. On the CPLID public dataset, the model increased the mAP50 by 9.6% compared to the baseline model.