长材表面标识字符识别技术研究  

Research on Character Recognition Technology of Long Material Surface Identification

作　　者：白金龙 王晓晨 徐言东 杨荃 郭强 肖雄 张勇军 BAI Jinlong;WANG Xiaochen;XU Yandong;YANG Quan;GUO Qiang;XIAO Xiong;ZHANG Yongjun

机构地区：[1]北京科技大学工程技术研究院,北京102206

出　　处：《中国重型装备》2024年第4期38-43,83,共7页CHINA HEAVY EQUIPMENT

基　　金：国家重点研发计划项目(2023YFB3712400)。

摘　　要：针对工业环境下长材表面标识字符常位于曲面上、训练样本少等常规场景文本识别方法难以处理的难点,提出一种针对工业环境下长材表面字符的识别技术。首先将二维图像通过自适应混合阈值二值化算法转化为适于流入检测算法的二值化图,并通过基于DBNet语义分割的字符定位算法及其对应的后处理算法定位、矫正字符所在区域后由单字符分割-识别方式完成识别。最后将所提出的方法进行对比验证,结果表明,所提出的针对长材表面字符识别各环节的改进相比于既有方法在识别准确率指标与直观效果层面均实现一定的提升,具备一定的工程应用可行性。Aiming at the difficulties of conventional scene text recognition methods,such as the mark characters on the surface of long products often located on the surface in the industrial environment,and the lack of training samples,this paper proposes a recognition strategy for the characters on the surface of long products in the industrial environment.Firstly,the two-dimensional image is transformed into a binary image suitable for the detection algorithm through the adaptive hybrid threshold binarization algorithm,and the character location algorithm based on DBNet semantic segmentation and its corresponding post-processing algorithm are used to locate and correct the located region,and then the single character segmentation recognition method is used to complete the recognition.Finally,the method proposed in this paper is compared and verified.The results show that the targeted improvement proposed in this paper for each link of long products surface character recognition has achieved a certain improvement in the recognition accuracy index and intuitive effect compared with the existing methods,and has a certain engineering application feasibility.

关 键 词：长材表面标识 混合阈值二值化 单字符识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]