| 注册
首页|期刊导航|现代电子技术|基于跨分量协同融合与多阶非局部通道注意力的绝缘子缺陷检测方法

基于跨分量协同融合与多阶非局部通道注意力的绝缘子缺陷检测方法

唐逸凡 余梅 陆林

现代电子技术2026,Vol.49Issue(6):160-167,8.
现代电子技术2026,Vol.49Issue(6):160-167,8.DOI:10.16652/j.issn.1004-373x.2026.06.024

基于跨分量协同融合与多阶非局部通道注意力的绝缘子缺陷检测方法

Method of insulator defect detection based on cross-component collaborative fusion and multi-order non-local channel attention

唐逸凡 1余梅 1陆林1

作者信息

  • 1. 三峡大学 水电工程智能视觉监测湖北省重点实验室,湖北 宜昌 443002||三峡大学 计算机与信息学院,湖北 宜昌 443002
  • 折叠

摘要

Abstract

In allusion to the high similarity between targets and background in insulator images,as well as the defect features of small targets are prone to being diluted due to downsampling and limited receptive fields,an insulator defect detection algorithm based on cross-component collaborative fusion and multi-order non-local channel attention is proposed.A cross-component collaborative fusion module is integrated into the backbone network,and the cross-domain fusion is conducted by means of fre-quency domain and spatial domain feature,to realize the multi-scale c and enhance the feature discrimination ability,thereby im-proving the recognition effect of subtle defect differences.In the neck network,a multi-order non-local channel attention mecha-nism is introduced to capture inter-channel correlations at multiple scales.In combination with non-local perception,it enhances the representation of small defect regions,suppresses feature dilution caused by downsampling,and then improves detection accuracy for small-scale defects.The experimental results show that the improved model can realize mAP@0.5 and mAP@0.5:0.95 of 79.7%and 38.6%,respectively,which are 3.6%and 3.8%higher than those of YOLOv8 benchmark model.The AP in the in-sulator defect category can reach 79.3%,and the frame rate can reach 60.2 f/s,which can meet the real-time detection require-ments of the power system and significantly improve the detection accuracy of insulator defects.

关键词

绝缘子/缺陷检测/跨分量协同融合模块/多阶非局部通道注意力/特征判别/频域信息

Key words

insulator/defect detection/cross-component collaborative interaction model/multi-order non-local channel attention mechanism/feature discrimination/frequency domain information

分类

信息技术与安全科学

引用本文复制引用

唐逸凡,余梅,陆林..基于跨分量协同融合与多阶非局部通道注意力的绝缘子缺陷检测方法[J].现代电子技术,2026,49(6):160-167,8.

基金项目

湖北省自然科学基金一般面上项目(2025AFB538) (2025AFB538)

现代电子技术

1004-373X

访问量0
|
下载量0
段落导航相关论文