基于ADFE-Net的航拍图像绝缘子缺陷检测  

Insulator Defect Detection in Aerial Images Based on ADFE-Net

作　　者：梁纲 马斌 饶宇飞 曹东升 栗晓政 钟封豪 LIANG Gang;MA Bin;RAO Yufei;CAO Dongsheng;LI Xiaozheng;ZHONG Fenghao(State Grid Henan Electric Power Research Institute,Zhengzhou 450003,China;State Grid Henan Electric Power Company,Zhengzhou 450000,China;State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources,North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网河南省电力公司电力科学研究院,河南郑州450003 [2]国网河南省电力公司,河南郑州450000 [3]新能源电力系统国家重点实验室(华北电力大学),北京102206

出　　处：《智慧电力》2025年第3期107-116,共10页Smart Power

基　　金：国家自然科学基金资助项目(62206095)。

摘　　要：为实现绝缘子缺陷的快速精准检测,提出一种基于自适应缺陷特征增强网络(ADFE-Net)的无锚框绝缘子缺陷检测方法。首先,在骨干网络中引入动态蛇形卷积(DSConv)提升细微缺陷特征的提取能力;其次,提出特征旋转交互模块(FRIM),以提升语义信息交互能力,进而抑制背景噪声干扰;最后,构建单输出加权双向特征金字塔网络(SWBFPN)实现多尺度特征融合,并增强对小尺寸缺陷的检测适应性。实验结果表明,ADFE-Net在检测精度及平均精度上达到了91.6%和90.4%,优于其它主流算法。该方法与无人机结合可以为电网系统的智能巡检提供参考。To achieve the rapid and accurate detection of insulator defects,the paper proposes an anchor-free insulator defect detection method based on ADFE-Net.Firstly,dynamic snake convolution(DSConv) is introduced into the backbone network to enhance the extraction capability of subtle defect features.Secondly,the feature rotation interaction module( FRIM) is proposed to improve semantic information interaction,thereby suppressing background noise interference.Finally,the scale-weighted bidirectional feature pyramid network(SWBFPN) is constructed to achieve multi-scale feature fusion and enhance the detection adaptability for small-sized defects.Experimental results show that ADFE-Net achieves 91.6% and 90.4% in detection accuracy and average precision,respectively,outperforming other mainstream algorithms.This method,combined with unmanned aerial vehicles(UAVs),can provide a reference for the intelligent inspection of power grid systems.

关 键 词：绝缘子 缺陷检测 动态蛇形卷积 金字塔网络 深度学习 

分 类 号：TM216[一般工业技术—材料科学与工程] TP391.41[电气工程—电工理论与新技术]