基于YOLOv5和U-net的指针式仪表读数识别方法  

Identification Method for Reading of Pointer Instruments Based on YOLOv5 and U-net

作　　者：金爱萍 袁亮 周德勤 杨奎刚 JIN Ai-ping;YUAN Liang;ZHOU De-qin;YANG Kui-gang(School of Mechanical Engineering,Xinjiang University,Urumqi 830017,China;School of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029,China;Beijing Yupont Electric Power Technology Co.,Ltd.,Beijing 100029,China)

机构地区：[1]新疆大学机械工程学院,新疆乌鲁木齐830017 [2]北京化工大学信息科学与技术学院,北京100029 [3]北京煜邦电力技术股份有限公司,北京100029

出　　处：《仪表技术与传感器》2022年第11期29-33,共5页Instrument Technique and Sensor

基　　金：新疆维吾尔自治区自然科学基金项目(2019D01C021)。

摘　　要：针对不同背景、距离条件下指针式仪表读数识别准确性低的问题,提出了一种结合YOLOv5和U-net的仪表读数识别方法。首先引入了YOLOv5算法检测并提取仪表区域;其次采用Hough梯度圆检测提取纯仪表盘和U-net分割算法准确分割出指针轮廓,并通过指针细化,加权最小二乘法拟合指针所在直线,根据定位坐标系获取指针方向和偏转角度,最后利用指针的偏转角度计算仪表读数。结果表明:提出的方法检测每张图像的平均耗时为64.511 ms,读数识别的成功率达97%,平均引用误差为0.577%,能快速并准确地识别出指针式仪表示数。Aiming at the problem of low accuracy of pointer instrument reading recognition under different background and distance conditions, a pointer instrument reading recognition method combining YOLOv5 and U-net was proposed.The YOLOv5 algorithm was introduced to detect and extract the instrument area.The Hough gradient circle detection was used to extract pure instrument disk.The U-net segmentation algorithm was utilized to segment the pointer contour, and the center of pointer was located by pointer refinement and weighted least squares.The pointer direction and deflection angle were obtained according to the positioning coordinate system, and the instrument reading was calculated by the pointer deflection angle.The results show that the average time consumed by the method to detect each image is 64.511 ms, the success rate of reading recognition reaches 97%,the average citation error is 0.577%,and it can identify the pointer instrument reading quickly and accurately.

关 键 词：指针式仪表 YOLOv5 U-net Hough梯度圆检测 加权最小二乘法 读数识别 

分 类 号：TH703[机械工程—仪器科学与技术] TP391[机械工程—精密仪器及机械]