基于机器学习和模板匹配的变电站仪表自动读数方法  

Automatic reading method of substation meters based on machine learning and template matching

作　　者：李汉巨 LI Hanju(China Southern Power Grid Digital Power Grid Research Institute Co.,Ltd,Guangzhou 510700)

机构地区：[1]南方电网数字电网研究院有限公司,广州510700

出　　处：《电气技术》2024年第1期61-66,共6页Electrical Engineering

摘　　要：针对变电站固定摄像头拍摄的各种类型的指针式仪表,提出一种自动读数方法。该方法包括模板制作、模板匹配、图像处理、表针识别和几何读数五个阶段。通过模板制作确定刻度值和角度的几何关系,使用模板匹配算法定位待读数的仪表盘位置,提取仪表盘部分的图像,并通过高斯滤波和伽马变换降低光照和阴影对表针识别的干扰。为提升复杂环境下图像二值化的效果,使用K均值聚类算法获取仪表盘图像的动态二值化阈值。为适应圆形和椭圆形变形的表盘,使用可变长度的线段拟合仪表盘二值化图像中的表针,获取表针的旋转角度,再结合主要刻度的角度和刻度值的对应关系,计算出表针角度对应的读数。实际应用结果表明,针对自然场景下变电站中的指针式仪表,该方法对光照、阴影、遮挡、倾斜、变形等干扰因素具有良好的鲁棒性,误差均小于最小刻度间隔,满足工程应用要求。An automatic reading method is proposed for various types of pointer instruments captured by fixed cameras in substations.This method consists of five stages:template making,template matching,image processing,needle recognition,and geometric reading.The geometric relationship between the scale value and the angle is determined by template making.The template matching algorithm is used to locate the position of the meter panel.The image of the meter panel is extracted.The interference of light and shadow on the needle recognition is reduced by Gaussian filter and gamma transform.In order to improve the effect of image binarization in complex environment,K-means clustering algorithm is used to obtain the dynamic binarization threshold of meter panel image.In order to adapt to the round and oval deformed dial,the line segment with variable length is used to fit the needle in the binary image of the meter panel to obtain the angle of the needle,and then the reading corresponding to the angle of the needle is calculated in combination with the corresponding relationship between the angle of the main scale and the scale value.The practical application results show that for the pointer meter in the substation,this method has good robustness to the interference factors such as illumination,shadow,occlusion,inclination and deformation,and the error is less than the minimum scale interval,which meets the engineering application requirements.

关 键 词：仪表读数 表针识别 模板匹配 K均值聚类 图像二值化 

分 类 号：TP181[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] TM63[电气工程—电力系统及自动化]