复杂背景下仪表信息的图像识别研究  

Image recognition research of instrument information based on complex background

作　　者：刘丽媛 刘宏展[1] LIU Liyuan;LIU Hongzhan(College of Information and Opto-electronic Science and Engineerings South China Normal University^Guangzhou 510006,China)

机构地区：[1]华南师范大学信息光电子科技学院,广州510006

出　　处：《激光杂志》2020年第4期66-69,共4页Laser Journal

基　　金：国家自然科学基金(No.61875057,61475049)。

摘　　要：复杂环境下仪器仪表的图像信号处理可广泛应用于水电表、燃气表、工厂锅炉表盘等机械工程领域,但是如何设计一种更高效、更简单的仪表信息识别技术,一直都是科研人员和工程师研究的热门话题。研究分析了三种可用于仪表信息识别的技术,并对比了其识别率和复杂度,对所采集的仪表图像的识别率分别为80.6%、49.2%、90.7%。实验结果表明,基于Python和OCR的仪表信息识别技术的识别效果是最佳的,对仪表图像进行预处理后得到待识别样本图,从而调用百度OCR的文字识别接口获取有用信息。相比于传统的基于嵌入式和OpenCV的识别技术,它不仅可以大大提高识别准确率至90%以上,而且还降低了复杂度。Image signal processing of instrumentation in complex environments can be widely used in mechanical engineering fields such as hydro-power meters,gas meters,and factory boiler dials,etc.But how to design a more efficient and simple instrument information identification technology has always been a hot topic of researchers and engineers.This paper analyzed three techniques that can be used for instrument information recognition,and compared its recognition rate and complexity.The recognition rates of the collected instrument images were 80.6%、49.2%and 90.7%,respectively.The experimental results show that the recognition effect of the instrument information recognition technology based on Python and OCR is the best.After preprocessing the instrument image,the sample image to be identified was obtained,and the text recognition interface of Baidu OCR was called to obtain useful information.Compared with traditional recognition technology based on embedded and OpenCV,it not only can greatly improve the recognition accuracy to more than 90%,but also reduce the complexity.

关 键 词：图像识别 数字图像处理 信息处理 光学字符识别 PYTHON 

分 类 号：TN911.73[电子电信—通信与信息系统]