基于旋转目标检测的指针仪表读数识别方法  

Reading recognition method of pointer instrument based on rotating target detection

作　　者：张尹人 邓春华 ZHANG Yin-ren;DENG Chun-hua(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;Institute of Big Data Science and Engineering,Wuhan University of Science and Technology,Wuhan 430065,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System,Wuhan University of Science and Technology,Wuhan 430065,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]武汉科技大学大数据科学与工程研究院,湖北武汉430065 [3]武汉科技大学智能信息处理与实时工业系统湖北省重点实验室,湖北武汉430065

出　　处：《计算机工程与设计》2023年第6期1804-1811,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61806150);湖北省科技厅计划基金项目(2018CFB195);湖北省教育厅科学技术研究计划青年人才基金项目(Q20181104);智能信息处理与实时工业系统湖北省重点实验室开放基金项目(znxx2018QN09);武汉科技大学国防预研基金项目(GF201814)。

摘　　要：为解决判断指针方向的问题,提出一个旋转目标检测网络R-YOLOv5。采用旋转矩形的形式定位指针的位置和角度方向,采用环形平滑标签的长边表示法区分矩形的方向,将角度回归预测问题转化为分类问题,设计合适的损失函数和旋转IoU计算方法,解决角度周期性对训练的影响。针对指针仪表数字刻度识别问题,提出一种空间聚类算法,能够鲁棒地识别刻度的数字。实验结果表明,所提指针仪表读数识别方法在误差允许范围内,识别准确率达93.9%。To solve the problem of judging the direction of the pointer,a rotating target detection network R-YOLOv5 was proposed.The form of rotating rectangle was used to locate the position and angle direction of the pointer,the long side representation of ring smooth label was used to distinguish the direction of the rectangle,the angle regression prediction problem was transformed into a classification problem,and an appropriate loss function and rotating IoU calculation method was designed to solve the influence of angle periodicity on training.Aiming at the problem of digital scale recognition of pointer instrument,a spatial clustering algorithm was proposed,which robustly recognized the number of scale.Experimental results show that within the allowable error range,the recognition accuracy of the proposed method is 93.9%.

关 键 词：指针方向 旋转目标检测网络 环形平滑标签 长边表示法 角度周期性 空间聚类 刻度识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]