[1]孙国栋,江亚杰,徐亮,等.BP网络预测阈值的仪表重影字符识别方法研究[J].郑州大学学报(工学版),2020,41(04):28-33.[doi:10.13705/j.issn.1671-6833.2020.04.011]
 Sun Guodong,Jiang Yajie,Xu Liang,et al.Study on Instrument ghosting character recognition method for predicting threshold by BP network[J].Journal of Zhengzhou University (Engineering Science),2020,41(04):28-33.[doi:10.13705/j.issn.1671-6833.2020.04.011]
点击复制

BP网络预测阈值的仪表重影字符识别方法研究()
分享到:

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
41卷
期数:
2020年04期
页码:
28-33
栏目:
出版日期:
2020-08-12

文章信息/Info

Title:
Study on Instrument ghosting character recognition method for predicting threshold by BP network
作者:
孙国栋江亚杰徐亮胡也席志远
湖北工业大学机械工程学院
Author(s):
Sun GuodongJiang YajieXu LiangHu YeXi Zhiyuan
School of Mechanical Engineering, Hubei University of Technology
关键词:
光照不均重影字符识别预测阈值LeNet-5BP神经网络
Keywords:
uneven lighting' target="_blank" rel="external">">uneven lightingghostingcharacter recognitionprediction thresholdsLeNet-5BP neural networks
DOI:
10.13705/j.issn.1671-6833.2020.04.011
文献标志码:
A
摘要:
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。实验表明,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%,单个字符识别时间只需要1.4ms。
Abstract:
In the process of instrument digital acquisition, there are many phenomena of uneven illumination and character double shadow, which lead to the difficulty of binarization and low recognition rate. A new binarization method was proposed. Before image binarization, the image needs to be preprocessed because of the poor image quality. For the phenomenon of uneven illumination, The color image correction method of nonlinear function was used.In view of the imaging ghosting, the images gray scale distribution statistics were taken as the input, and the adaptive binarization global threshold was 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, so as to achieve the separating the ghosting. At the same time, the improved LeNet-5 network was used to recognize the single character after segmentation. The experimental results show that the proposed binarization method is better than the classical method, and the improved LeNet-5 can satisfy the instrument character recognition after segmentation, with the recognition rate of 98.94%, and the single character recognition time of only 1.4ms.
更新日期/Last Update: 2020-10-06