基于YOLOv5算法的无人机巡检图像中绝缘子单目测距方法  

Monocular Distance Measurement Method for Insulator in Unmanned Aerial Vehicle Inspection Images Based on YOLOv5 Algorithm

作　　者：陈彬[1] 刘华洲 李勃铖 贾燕峰 郭昊 孙君录 CHEN Bin;LIU Huazhou;LI Bocheng;JIA Yanfeng;GUO Hao;SUN Junlu(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China;Ultra High Voltage Transmission Company of State Grid Hunan Electric Power Co.,Ltd.,Changsha 410100,China;Sanmenxia Power Supply Company,State Grid Henan Electric Power Company,Sanmenxia 472000,China;Pingdingshan Power Supply Company,State Grid Henan Electric Power Company,Pingdingshan 467002,China)

机构地区：[1]三峡大学电气与新能源学院,湖北宜昌443002 [2]国网湖南省电力有限公司超高压输电公司,湖南长沙410100 [3]国网河南省电力公司三门峡供电公司,河南三门峡472000 [4]国网河南省电力公司平顶山供电公司,河南平顶山467002

出　　处：《无线电工程》2024年第6期1421-1430,共10页Radio Engineering

摘　　要：为了提升无人机安全飞行性,提升线路运维能力,基于YOLOv5深度学习算法,提出了一种无人机单目测距方法,用于实现一种精确的辅助距离测量功能。对现有的开源绝缘子数据集进行了扩充和标定,采用YOLOv5s模型进行了训练、验证和测试,建立了绝缘子串的测距模型,设计代码加入到检测模块中,实现对绝缘子的识别和测距。实验结果表明,该模型能够准确识别输电线路不同类型的绝缘子并进行精确测距,平均测距误差为4.76%;对复合绝缘子的识别和测距效果最佳,改变俯仰角和相对距离,最大测距误差为6%;在不同天气条件下,亮度变化越大,对绝缘子的识别测距误差也越大,亮度增加至100%时,误差最高可达到12.3%。测距所需平均时间为0.2984 s,可以实现高效、高精度测距,为无人机巡检安全距离测量提供支持。In order to enhance the security of UAV flights and improve line operation and maintenance capabilities,a UAV monocular ranging method is proposed for implementing an accurate auxiliary distance measurement function based on the YOLOv5 deep learning algorithm.The existing open source insulator dataset is expanded and calibrated,then the YOLOv5s model is trained,validated and tested to build a distance measurement model for insulator strings,and the design code is added to the detection module to realize the identification and ranging of insulators.The experimental results show that the model can accurately identify different types of insulators on transmission lines and perform accurate ranging,with an average ranging error of 4.76%;the model is most effective in identifying and ranging composite insulators,and the maximum ranging error is 6%when changing the pitch angle and relative distance;the greater the change in brightness under different weather conditions,the greater the error in identifying and ranging insulators,when the brightness increases to 100%,the error can reach up to 12.3%.The average time required for distance measurement is 0.2984 s.The proposed method enables efficient and highly accurate distance measurement and provides support for safe distance measurement for UAV inspection.

关 键 词：安全距离 绝缘子 深度学习 YOLOv5 图像识别 单目测距 

分 类 号：TM755[电气工程—电力系统及自动化] TP391[自动化与计算机技术—计算机应用技术]