改进神经网络算法在车牌识别中的应用  

Application of License Plate Recognition Based on Improved Neural Network

作　　者：丁伟[1] 

机构地区：[1]南阳理工学院,河南南阳473004

出　　处：《计算机仿真》2011年第8期359-362,366,共5页Computer Simulation

摘　　要：研究现代智能交通管理中的车牌准确识别问题。由于车牌图像存在模糊不清、倾斜,分割后字符图像笔画粗细不均、断续不完整等特殊性,导致传统车牌识别算法识别速度慢、识别正确率低,不能适应车牌识别的实时性要求。为了提高车牌识别正确率,提出一种BP神经网络的车牌识别算法。该算法首先对车牌字符图像进行归一化处理,消除图像中无用信息,然后对车牌字符进行特征提取,消除笔画粗细不均、断续不完整等影响,再将提取车牌字符特征输入到BP神经网络进行学习和识别,并采用动量因子和自适应学习速率对BP神经网络进行改进,加快其收敛速度,从而提高识别的实时性。仿真结果表明,改进用BP神经网络提高了车牌识别正确率和识别速度,缩短了识别时间,适合于实时性强的智能交通管理系统应用。License plate recognition is core technology in the modern intelligent transportation management.The BP neural network has slow convergence speed,and is easy to fall into the local optimal problem.A license plate recognition algorithm is proposed based on improved BP neural network.Firstly the vehicle license plate characters are normalized,some useless information is eliminates,and character features are extracted.Then the BP neural network is improved by momentum factor and adaptive learning speed,and its convergence rate jis accelerated.Finally the extracted character features are input into the BP neural network for training and recognition.Simulation experiments show that compared with other license plate recognition algorithm,the improved BP neural network improves license plate recognition accuracy and recognition efficiency,and it is very suitable for real-time modern intelligent traffic management.

关 键 词：车牌识别 字符识别 神经网络 特征提取 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]