BP网络预测阈值的仪表重影字符识别方法研究  

Study on Instrument Ghosting Character Recognition Method for Predicting Binarization Threshold by BP Network

作　　者：孙国栋[1] 江亚杰 徐亮 胡也 席志远 SUN Guodong;JIANG Yajie;XU Liang;HU Ye;XI Zhiyuan(School of Mechanical Engineering,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068

出　　处：《郑州大学学报（工学版）》2020年第4期28-33,共6页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(51775177,51675166);湖北省自然科学基金资助项目(2018CFB276)。

摘　　要：针对仪表数字获取过程中多出现光照不均匀和字符重影现象,导致二值化困难、识别率低等问题,提出了一种新的二值化方法。在对图像二值化之前,由于图像质量不佳,首先需要对图像进行预处理。针对光照不均现象,使用了非线性函数彩色图像校正方法。针对重影现象,以图像的灰度级分布统计量作为输入,自适应二值化全局阈值作为标签训练BP神经网络预测模型,使用训练好的BP网络对图像全局阈值进行预测并二值化,达到分离重影的目的。同时,采用改进LeNet-5网络对分割后的单个字符进行识别。结果表明,提出的二值化方法效果优于经典方法,改进的LeNet-5能够满足分割后的仪表字符识别,其识别率能达到98.94%,分割后单个字符识别时间只需要0.0014 s。During instrument digital image acquisition,there were many phenomena of uneven illumination and character double shadow,which led to the difficulty of binarization and low recognition rate.A new binarization method is proposed.Before image binarization,the image would be preprocessed because of the poor image quality.Due to uneven illumination,the color image correction method based on nonlinear function was used.In view of the imaging ghosting,the image gray scale distribution statistics were taken as the input,and the adaptive binarization global threshold is used as the label of prediction model to train BP neural network.The trained BP network was used to predict the global threshold and binarize the image,in order to achieve the separating the ghosting.At the same time,the improved LeNet-5 network was adopted to recognize the single character after segmentation.The experimental results showed that the proposed binarization method was better than the classical methods,and the improved LeNet-5 could satisfy the instrument character recognition after segmentation,with the recognition rate of 98.94%,and the single character recognition time of only 0.0014s.

关 键 词：光照不均 重影 字符识别 预测阈值 LeNet-5 BP神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]