基于Faster R-CNN的变电站指针式仪表识别技术研究  

Research on pointer meter recognition technology for substation based on Faster R-CNN

作　　者：张永斌 李东 崔杨 王海帆 ZHANG Yongbin;LI Dong;CUI Yang;WANG Haifan(Nanyang Power Supply Company,Henan Electric Power Company,State Grid,Nanyang 473000,China;Zhengzhou University of Light Industry,Zhengzhou 450000,China)

机构地区：[1]国网河南省电力公司南阳供电公司,河南南阳473000 [2]郑州轻工业大学,河南郑州450000

出　　处：《电气应用》2024年第9期37-43,共7页Electrotechnical Application

摘　　要：针对变电站仪表读数准确性差和效率低的问题,提出了一种基于Faster R-CNN的目标检测算法和EmguCV图像处理库结合的方法。首先引入了Faster R-CNN两阶段目标检测算法,对仪表盘进行检测识别和提取。其次,采用随机采样圆检测进行圆形表盘定位,使用Laplace算子处理图像提取主刻度线并结合环向模板匹配法进行仪表指针定位。最后,通过指针偏移角度来计算最终示数。通过仿真实验表明:提出的方法对仪表的识别率达到了99.38%,自动读数系统平均误差小于0.2%,对每张图片的处理时间约为35 ms。Aiming at the low accuracy and efficiency of meter reading in substation,a method based on Faster R-CNN and EmguCV image processing library is proposed.Firstly,Faster R-CNN two-stage object detection algorithm is introduced to detect,identify and extract the dashboard..secondly,the random sampling circle detection was adopted for circular dial positioning,Laplace operator was used to process the main scale line of image extraction and the Circumferential Template Matching for instrument pointer positioning.finally,the final indicator was calculated by the offset Angle of the pointer.The simulation results show that the recognition rate of the instrument is 99.38%,the average error of the automatic reading system is less than 0.2%,and the processing time of each image is about 35 ms.

关 键 词：仪表自动读数 两阶段目标检测算法 图像处理库 随机采样圆检测 环向模板匹配法 

分 类 号：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] TM63[电气工程—电力系统及自动化]