融合双域特征的绝缘子缺陷检测算法
DOI:
CSTR:
作者:
作者单位:

1.陕西科技大学 电气与控制工程学院;2.陕西科技大学 电子信息与人工智能学院

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金项目(61871260)


Insulator Defect Detection Algorithm Based on Dual-Domain Feature Fusion
Author:
Affiliation:

1.School of Electrical and Control Engineering,Shaanxi University of Science and Technology,Xi’an;2.School of Electrical Information and Artificial Intelligence, Shaanxi University of Science and Technology,Xi’an

Fund Project:

The National Natural Science Foundation of China (61871260)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    在无人机航拍输电线路巡检场景中,针对复杂环境干扰导致的绝缘子缺陷检测精度不高、误检漏检等问题,以RT-DETR为基础,提出了一种融合双域特征的绝缘子缺陷检测算法。在主干网络中设计空频特征融合模块(FreqSF),通过融合空间域与频率域信息,增强了绝缘子缺陷细节纹理特征的表达能力;引入高低频注意力机制(HiLo)替换多头自注意力机制(MHSA),将特征图中的高频与低频特征分开处理,以强化对局部细节特征的提取;采用三重特征编码器(TFE)与尺度序列特征融合模块(SSFF),以将深层语义信息与浅层细节信息进行深度融合。实验结果表明,所提算法相较于主流算法取得了较好的检测效果,其mAP@0.5达到0.974,相较于基线算法提高了1.5%,精确率和召回率分别提高了1.2%和1.1%。

    Abstract:

    To address the issues of low detection accuracy and frequent false positives and false negatives caused by complex environmental interferences during unmanned aerial vehicle inspections of power transmission lines, this paper proposes a dual-domain feature fusion algorithm for insulator defect detection based on RT-DETR. First, a Spatial-Frequency Feature Fusion (FreqSF) module is designed within the backbone network. By integrating spatial and frequency domain information, this module significantly enhances the feature representation capability for fine textures of insulator defects. Second, a High-Low Frequency Attention (HiLo) mechanism is introduced to replace the traditional Multi-Head Self-Attention (MHSA). By processing high-frequency and low-frequency features separately, it strengthens the extraction of local detail features. Finally, a Triple Feature Encoder (TFE) and a Scale Sequence Feature Fusion (SSFF) module are employed to achieve the deep fusion of high-level semantic information and low-level detail information. Experimental results demonstrate that the proposed algorithm outperforms mainstream detection models. Specifically, it achieves a mAP@0.5 of 0.974, yielding an increase of 1.5% over the baseline algorithm, while the precision and recall are improved by 1.2% and 1.1%, respectively.

    参考文献
    相似文献
    引证文献
引用本文
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2026-04-12
  • 最后修改日期:2026-06-08
  • 录用日期:2026-06-17
  • 在线发布日期:
  • 出版日期:
文章二维码