首页|输电线路金具巡检图像倾斜缺陷检测技术研究

输电线路金具巡检图像倾斜缺陷检测技术研究

扫码查看
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.
Research on Tilted Defect Detection Technology of Transmission Line Fitting Inspection Image
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.

Multi-scale Residual Networkrotate anchor boxgrading ring tilted detectionclamp offset defectinsulator tilt detection

王万国、孟弘毅、许荣浩、杨月琛、李振宇、王正

展开 >

国网智能科技股份有限公司,山东 济南 250000

山东省实验中学,山东 济南 250000

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

国家电网有限公司总部管理科技项目

5108-202218280A-2-249-XG

2024

现代信息科技
广东省电子学会

现代信息科技

ISSN:2096-4706
年,卷(期):2024.8(16)
  • 10