摘要
为提高手持式数字万用表校准系统的自动化水平,本文通过研究图像灰度阈值法、传统方法的目标分类和深度学习三种不同类型的字符识别技术,提出了两种基于机器视觉的数字识别方案。测试结果显示,两种方案的字符识别准确率均可达到99.8%,但其在硬件资源占比、编程难易程度上二者还存在显著差异。该机器视觉字符识别功能的成功开发与应用,可为更多无程控通信接口的计量测试设备,及一些不适于人工作业的危险计量工作环境进行类似的数字识别提供借鉴。
To improve the automation level of the handheld digital multimeter calibration system,two methods of digital recognition schemes based on machine vision were proposed by studying three different types of character recognition technology,namely image gray threshold method,target classification by traditional methods and deep learning.The test result shows that although the character recognition accuracy of two schemes can both reach 99.8%,there are significant differences in proportion of hardware resources and degree of programming difficulty.After this machine vision character recognition function has been successfully developed,it will provide useful reference for more metering test equipment without program-controlled communication interface,or similar digital recognition in some dangerous metering working environment which is not suitable for manual operation.
作者
彭诚
丁蔚
侯旭玮
李军
PENG Cheng;DING Wei;HOU Xu-Wei;LI Jun(Beijing Oriental Institute of Measurement and Test,Beijing 100086,China)
出处
《宇航计测技术》
CSCD
2020年第1期61-66,共6页
Journal of Astronautic Metrology and Measurement
关键词
机器视觉
数字万用表
自动化
字符识别
数字识别
深度学习
Machine vision
Digital multimeter
Automation
Character recognition
Digital recognition
Deep learning
