基于OCR的中文债券图表数据检测和文本识别  

OCR Based Chinese Bond Chart Data Detection and Text Recognition

作　　者：张宁静 袁书培 吴海龙[1] Zhang Ningjing;Yuan Shupei;Wu Hailong(College of Computer Science,University of South China,Hengyang 421000)

机构地区：[1]南华大学计算机学院,衡阳421000

出　　处：《现代计算机》2021年第30期73-81,共9页Modern Computer

基　　金：湖南省大学生创新创业训练计划项目(202110555051)。

摘　　要：针对金融业务场景下海量债券图表的人工处理效率低、处理容错性低、数据安全性高等问题,本文主要基于OCR技术设计了应用于中文债券图表数据检测和文本识别的解决方案,并且基于数据安全性及实际生产要求,采用离线安卓移动端作为算法使用平台。本文在原有的DBNet和CRNN网络的基础上对算法做出适应场景的改进,实验结果表明,本文所提算法在检测识别速度下降不明显和实现离线移动端部署的前提下检测识别效果更好,具有极大的工程应用价值。For financial business scenarios mass of bond graph,low efficiency of manual processing,processing low fault tolerance and high data security problem,this paper is mainly based on OCR technology to design the utilized bond graph data detection and text recognition solutions,and based on the requirement of data security and the actual production,using offline android mobile terminal as algorithm using the platform.Based on the DBNet and CRNN network,this paper improves the algorithm to adapt to the scene.The ex⁃perimental results show that the algorithm proposed in this paper has better detection and identification effect under the premise that the detection and identification speed is not obvious and offline mobile deployment is realized,which has great engineering application value.

关 键 词：OCR光学字符识别技术 文字检测 文字识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]