基于机器视觉的工业机床信息识别方法  

Machine Vision Based Information Collection Method of Industrial Machine

作　　者：柏森 孙长江[1,2] 江录春 陈根良[1,2] 王皓[1,2] BAI Sen;SUN Changjiang;JIANG Luchun;CHEN Genliang;WANG Hao(Shanghai Key Laboratory of Digital Manufacture for Thn-walled Structures,Shanghai Jiao Tong University,Shanghai 200240,China;Key Laboratory of Mechanical System and Vibration,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学上海市复杂薄板结构数字化制造重点实验室,上海200240 [2]上海交通大学弗劳恩霍夫协会智能制造项目中心,上海201306

出　　处：《机械设计与研究》2022年第1期78-84,共7页Machine Design And Research

基　　金：国家科技重大专项(2019YFA0709001);国防基础科研项目(JCKY2017203 A007)。

摘　　要：针对目前部分工厂急需对设备进行数字化升级,而国外机床厂家不开放机床内部数据接口,需要.通过机器视觉、外接传感器等方式采集机床工作信息的问题,本论文提出了一种基于二维视觉进行图像采集并降噪,使用卷积神经网络进行信息识别的机床信息采集方法。首先,基于投影变换原理确定机床面板范围,算法不受光照影响;再提出一种针对工业现场条件变化导致字符识别困难的问题的面板噪声滤波方法对采集图像进行预处理;然后,基于卷积神经网络,对不同类型的字符统一识别;最终,搭建了视觉识别系统并在某军工厂进行部署,实现了稳定的工业化运行。Aiming at the problem that some factories urgently need to digitally upgrade their equipment,the fact that foreign machine tool manufacturers do not open the internal data interface of the machine tool,and the need to collect machine tool work information through machine vision,external sensors,etc.,this paper proposes a two-dimensional vision-based image collection and noise reduction method using a convolutional neural network to identify machine tool information.Firstly,the range of the machine tool panel is determined based on the principle of projection transformation,and the algorithm is not affected by lighting conditions;further a panel noise filtering method is proposed to deal with the problem of difficult character recognition caused by changes in industrial field conditions to preprocess the collected images;then,based on convolutional neural network,recognition of different types of characters is unified;finally,a visual recognition system was built and deployed in a military factory,achieving stable industrial operation.

关 键 词：工业机床 字符识别 机器视觉 图像降噪 仿射变换 图像分类 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]