基于OCR技术的复杂背景下工件标识字符识别方法  

Recognition of Workpiece Identifier on Complex Background Using OCR Technology

作　　者：刘基 赵志诚[1] 王晓东 LIU Ji;ZHAO Zhi-cheng;WANG Xiao-dong(School of Electronic Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,Shanxi,China)

机构地区：[1]太原科技大学电子与信息工程学院,山西太原030024

出　　处：《铸造》2021年第7期855-860,共6页Foundry

基　　金：山西省重点研发计划项目(201803D121025);山西省留学人员科技活动择优资助项目(2020005)。

摘　　要：针对铸造工艺过程中白模挂涂、铸件补漆后工件标识字符模糊不清、背景复杂,从而导致识别难度增加的问题,提出了一种基于卷积循环神经网络与连接时域分类的端对端光学字符识别模型。该模型在卷积层基于VGGNet16进行了简化和改进,又利用旋转、加噪、调整亮度和对比度的数据增广方法解决了样本数量少的问题。根据试验对比选择合适的模型参数,实现了复杂背景下铸造工件标识字符的识别。结果表明,改进后的字符识别模型稳定、识别率高,对相似字符具有较好的鲁棒性。The workpiece characters is blurred and has a complex background after the white pattern and the casting are painted,which leads to the difficulty of identifying the workpiece characters is gradually increasing.In order to solve the problem,the author proposes an end-to-end optical character recognition(OCR)model based on convolutional recurrent neural network(CRNN)and connectionist temporal classification(CTC),and the model is simplified and improved based on VGGNet16 in small amount of data by augmented method of rotate,adding noise,adjusting brightness and contrast.According to the experimental comparison,the appropriate model parameters are selected to recognize the identification characters of cast workpiece in complex background.The results show that this improved model is stable and has high recognition rate,and has great robustness for similar characters.

关 键 词：铸造工件 工件标识符 光学字符识别 卷积循环神经网络 连接时域分类 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]