输电线路金具巡检图像倾斜缺陷检测技术研究  

Research on Tilted Defect Detection Technology of Transmission Line Fitting Inspection Image

作　　者：王万国 孟弘毅 许荣浩 杨月琛 李振宇 王正 WANG Wanguo;MENG Hongyi;XU Ronghao;YANG Yuechen;LI Zhenyu;WANG Zheng(State Grid Intelligence Technology Co.,Ltd.,Ji'nan 250000,China;Shandong Experimental High School,Ji'nan 250000,China)

机构地区：[1]国网智能科技股份有限公司,山东济南250000 [2]山东省实验中学,山东济南250000

出　　处：《现代信息科技》2024年第16期113-117,共5页Modern Information Technology

基　　金：国家电网有限公司总部管理科技项目(5108-202218280A-2-249-XG)。

摘　　要：针对当前输电线路无人机巡检图像中金具倾斜缺陷存在角度量化困难、判定主观性强的问题,提出融合多尺度残差网络的金具倾斜类缺陷检测技术。通过中心点坐标和旋转角结合的方式表示有向边界框,实现设备倾斜角度的有效量化;利用角度回归提高倾斜锚框检测灵敏度,实现输电线路倾斜类缺陷的识别和精确定位。经现场验证,线夹偏移类缺陷检测准确率由86.2%提升至95.02%,均压环倾斜类缺陷检测准确率由87.76%提升至98.11%,绝缘子倾斜类缺陷检测准确率由85.34%提升至96.12%,满足现场应用需求。Aiming at the problems of difficult angle quantification and strong subjectivity in determination of tilted defect of fitting in the current UAV inspection image of transmission line,the tilted defect detection technology of fitting tilted class is proposed by integrating Multi-scale Residual Network.Through the combination of center point coordinates and rotation angle to represent the directed bounding box,the effective quantification of the tilted angle of the equipment is realized.The angle regression is used to improve the detection sensitivity of the tilted anchor box,and to realize the identification and precise positioning of the tilted defect in transmission line.After on-site verification,the detection accuracy of clamp offset defect is increased from 86.2%to 95.02%,the detection accuracy of grading ring tilted defect is increased from 87.76%to 98.11%,and the detection accuracy of insulator tilted defect is increased from 85.34%to 96.12%,which meets the requirements of on-site application.

关 键 词：多尺度残差网络 旋转锚框 均压环倾斜检测 线夹偏移缺陷 绝缘子倾斜检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]