查询结果:   张国云,向灿群,吴健辉,郭龙源,涂兵.基于改进BP网络的车牌字符识别方法研究[J].计算机应用与软件,2017,34(4):243 - 248.
中文标题
基于改进BP网络的车牌字符识别方法研究
发表栏目
图像处理与应用
摘要点击数
746
英文标题
RESEARCH ON LICENSE PLATE CHARACTER RECOGNITION METHOD BASED ON IMPROVED BP NEURAL NETWORK
作 者
张国云 向灿群 吴健辉 郭龙源 涂兵 Zhang Guoyun Xiang Canqun Wu Jianhui Guo Longyuan Tu Bing
作者单位
湖南理工学院信息与通信工程学院 湖南 岳阳 414006 复杂系统优化与控制湖南省普通高等学校重点实验室 湖南 岳阳 414006    
英文单位
College of Information and Communication Engineering,Hunan Institute of Science and Technology,Yueyang 414006,Hunan,China Key Laboratory of Optimization and Control for Complex Systems,Hunan Institute of Science and Technology,Yueyang 414006,Hunan,China    
关键词
改进BP网络 车牌 字符识别 全参数自动调整
Keywords
Improved BP neural network License plate Character recognition All parameter automatic adjustment
基金项目
湖南省教育厅产业化培育项目(13CY021);湖南省教育厅开放基金项目(15K051)
作者资料
张国云,教授,主研领域:机器视觉。向灿群,硕士生。吴健辉,副教授。郭龙源,副教授。涂兵,讲师。 。
文章摘要
针对传统BP算法在车牌字符识别速度较慢和识别准确率较低的问题,提出一种改进的BP网络车牌字符识别方法。通过对BP算法的输入特征数优化,在不降低识别精度的情况下精简了输入层节点数,提升了识别速度。改进后的BP算法采用全参数自动调整,引入自适应学习率、动量因子、坡度因子,增加了BP算法的识别精度;同时通过更好的利用车牌字符特征和BP网络特征,降低了算法结构的复杂性,增强了算法的鲁棒性。实验结果表明,该算法在实际采集的自建整副车牌数据集上的识别率上比传统BP神经网络车牌识别算法提高近6.5%;在识别速度上提高近1.3 s。
Abstract
Aiming at the problem that the traditional BP algorithm is slow in the recognition speed of the license plate and the recognition accuracy is low, an improved BP neural network license plate character recognition method is proposed. By optimizing the input feature number of BP algorithm, the number of nodes in the input layer is reduced and the recognition speed is improved without reducing the recognition accuracy. The improved BP algorithm adopts the all parameter automatic adjustment and introduces adaptive learning rate, momentum factor and slope steepness factor, which increases the recognition accuracy of BP algorithm. At the same time, through better use of the license plate character features and BP network features, it reduced the complexity of the algorithm structure, and enhanced the robustness of the algorithm. The experimental results show that the recognition rate of the algorithm is 6.5% higher than that of the traditional BP neural network license plate recognition algorithm based on the self-built license plate data set, and the recognition speed is improved by 1.3 s.
下载PDF全文