深度学习目标检测算法在架空输电线路绝缘子缺陷检测中的应用研究综述  

Review of Application Research of Deep Learning Object Detection Algorithms in Insulator Defect Detection of Overhead Transmission Lines

作　　者：刘开培[1] 李博强 秦亮[1] 李强 赵峰 王秋琳 许中平 余金沄 LIU Kaipei;LI Boqiang;QIN Liang;LI Qiang;ZHAO Feng;WANG Qiulin;XU Zhongping;YU Jinyun(School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China;State Grid Information&Telecommunication Group Co.,Ltd.,Beijing 102211,China;Fujian Yirong Information Technology Co.,Ltd.,Fuzhou 350003,China;Beijing SGITG-Accenture Information Technology Co.,Ltd.,Beijing 100052,China)

机构地区：[1]武汉大学电气与自动化学院,武汉430072 [2]国网信息通信产业集团有限公司,北京102211 [3]福建亿榕信息技术有限公司,福州350003 [4]北京国网信通埃森哲信息技术有限公司,北京100052

出　　处：《高电压技术》2023年第9期3584-3595,共12页High Voltage Engineering

基　　金：国家重点研发计划(2020YFB0905900)。

摘　　要：传统架空输电线路绝缘子缺陷检测一般通过人工巡检方式进行。架空输电线路的数量增长使巡检规模更加庞大、巡检环境更加复杂,放大了传统绝缘子缺陷检测方法人力成本高、检测效率低的不足。无人机(unmanned aerial vehicle,UAV)等新型巡线方式依靠深度学习目标检测算法识别架空输电线路绝缘子缺陷,能够有效应对人工巡检的不足,是绝缘子缺陷检测的发展趋势。鉴于此,围绕架空输电线路绝缘子缺陷检测场景,首先梳理常用的深度学习目标检测算法,比较不同算法的检测策略、检测精度与检测速度;然后结合云–边–端协同架构说明算法的改进需求与相应改进方法;最后针对现有绝缘子检测方面的不足,展望了输电线路绝缘子中多类型缺陷的识别问题,并在这一研究趋势下进一步探讨了模型边缘端轻量化与针对小样本数据下的算法研究价值。Traditional defect detection for overhead transmission line insulator is generally carried out by manual inspection.The increase in the number of overhead transmission lines has made larger scale and more complex environment of the inspection,which amplifies the shortcomings of traditional insulator defect detection methods with high labor costs and low detection efficiency.New line inspection methods such as unmanned aerial vehicle(UAV)rely on deep learning object detection algorithms to identify insulator defects in overhead transmission lines,which effectively deals with shortcomings of manual inspection and becomes the development trend of insulator defect detection.Therefore,focusing on defect detection scenario of overhead transmission line insulator,we firstly sorted out the commonly used deep learning object detection algorithms,and compared the detection strategies,detection accuracy and detection speed of different algorithms.Then,combined with the cloud-edge-end collaborative architecture,the improvement requirements of the algorithms and corresponding improvement methods of the algorithms were explained.Finally,in response to the shortcomings of existing insulator detection,the identification of multiple types of defects in transmission line insulators is prospected,and under this research trend,the value of model edge lightweight and algorithm research for small sample data is further explored.

关 键 词：架空输电线路绝缘子 缺陷检测 无人机 深度学习目标检测算法 云–边–端协同架构 

分 类 号：TM75[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]