基于图像深度学习的电力工程巡视识别算法设计  

Design of patrol recognition algorithm for power engineering based on image depth learning

作　　者：刘军 杨亚萍 王洪亮 LIU Jun;YANG Yaping;WANG Hongliang(State Grid Gansu Electric Power Company,Lanzhou 730070,China;State Grid Longnan Electric Power Supply Company,Wudu 746000,China)

机构地区：[1]国网甘肃省电力公司,甘肃兰州730070 [2]国网陇南供电公司,甘肃武都746000

出　　处：《电子设计工程》2024年第6期180-184,共5页Electronic Design Engineering

基　　金：甘肃陇南供电公司2022年管理创新项目(D22FZ2712008)。

摘　　要：针对电力工程线路图像信息利用程度低的问题,基于图像深度学习技术,文中提出了电力线路档距智能识别与审计算法。该算法利用维纳滤波进行电力线路图像的预处理,采用极线矫正和立体匹配完成实际空间点在双目图像中的映射匹配。同时使用Sobel算子进行边缘检测,并通过三维坐标转换实现电力线路档距的智能识别与审计。仿真算例结果表明,所提算法的档距识别平均误差仅为3.4%,相比于单目测距算法具有更高的识别准确度。且在实际电力工程应用中,该算法对档距不合格线路的识别率高达85%,远大于单目测距法的40%,故可有效提升电力工程线路审计的智能化水平。Aiming at the problem of low utilization of power line image information,based on image depth learning technology,an intelligent recognition and audit algorithm of power line span is proposed in this paper.The algorithm uses Wiener filter to preprocess the power line image,uses epipolar correction and stereo matching to realize the mapping and matching of actual space points in the binocular image,and then uses Sobel operator to realize edge detection.Through three⁃dimensional coordinate tran⁃sformation,it realizes the intelligent recognition and audit of power line span.The simulation results show that the span recognition error of the proposed algorithm is only 3.4%,which has higher recognition accuracy than the monocular ranging algorithm.In the actual application of power engineering,the recognition rate of the algorithm for the unqualified line of span is as high as 85%,which is much higher than the recognition rate of monocular ranging method of 40%.It can effectively improve the intelligent level of power engineering line audit.

关 键 词：双目测距 图像处理 档距审计 电力工程数据分析 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TN99[自动化与计算机技术—控制科学与工程]