基于卷积神经网络的字轮式仪表双半字符识别  

Double half-character recognition for wheel-type meter based on convolutional neural network

作　　者：王望 徐望明[1,2,3] 伍世虔 闫富海 Wang Wang;Xu Wangming;Wu Shiqian;Yan Fuhai(College of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan 430081,China;Institute of Robotics and Intelligent Systems, Wuhan University of Science and Technology, Wuhan 430081, China;Engineering Research Center for Metallurgical Automation and Detecting Technology of Ministry of Education, Wuhan University of Science and Technology, Wuhan 430081, China)

机构地区：[1]武汉科技大学信息科学与工程学院,湖北武汉430081 [2]武汉科技大学机器人与智能系统研究院,湖北武汉430081 [3]武汉科技大学冶金自动化与检测技术教育部工程研究中心,湖北武汉430081

出　　处：《武汉科技大学学报》2021年第1期68-73,共6页Journal of Wuhan University of Science and Technology

基　　金：国家自然科学基金资助项目(61775172,51805386);湖北省教育厅科研计划项目(D20191104).

摘　　要：针对远程自动抄表系统中字轮式表盘出现双半字符导致识别困难的问题,提出一种基于卷积神经网络(CNN)的双半字符识别方法。对采集到的表盘图像,首先通过图像预处理和投影算法分割出完整字符和双半字符。鉴于实际图像中双半字符样本较少,不足以训练深度网络,该方法通过收集相邻数字的完整字符图像进行对齐拼接,再从中随机截取子图像从而建立一个可用的双半字符训练集,并训练好专门设计的CNN模型。以燃气自动抄表系统读数为例,在生成的双半字符测试集和真实的双半字符测试集上进行实验,结果表明,与传统方法相比,本文方法有效提高了双半字符识别准确率。It is difficult to recognize double half-character images on the dials of wheel-type meters in remote automatic meter reading system,so a noval method for double half-character recognition based on convolutional neural network(CNN)was proposed.Firstly,image preprocessing and projection algorithms were applied to segment the collected dial images for complete characters and double half-characters.Since double half-character samples in the real images were not enough to train a deep network,a usable training dataset of double half-character images was established from the complete character samples by aligning and stitching the images of adjacent numbers and then cropping on them randomly,thus a specially designed CNN model could be trained.Taking the automatic reading of gas meter as an example,experiments were carried out on a generated double half-character testing dataset and a real one.The results indicate that the proposed method can effectively improve the recognition accuracy for double half character images when compared with the traditional methods.

关 键 词：远程自动抄表 双半字符识别 字轮式仪表 卷积神经网络 图像预处理 投影直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]