改进深度学习的车牌字符识别技术  

License plate recognition technology by new depth learning

作　　者：郑祥盘 王兆权[1] 宋国进 ZHENG Xiangpan;WANG Zhaoquan;SONG Guojin(Fujian Provincial Key Laboratory of Advanced Motion Control,Minjiang University,Fuzhou,Fujian 350108,China;College of Mechanical Engineering and Automation,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]闽江学院福建省先进运动控制重点实验室,福建福州350108 [2]福州大学机械工程及自动化学院,福建福州350108

出　　处：《福州大学学报（自然科学版）》2021年第3期316-322,共7页Journal of Fuzhou University(Natural Science Edition)

基　　金：福建省科技厅引导性项目(2019H0028);闽江学院人才引进项目(MJY19029);闽江学院实验实训中心专项(MJUS2019A006)。

摘　　要：针对传统车牌字符检测方法存在效率低、可靠性差的情况,提出应用Haar级联检测结合深度学习方法的卷积神经网络车牌字符识别法.首先采用Haar级联分类器提取出图片中车牌的位置,通过灰度、阈值、腐蚀、膨胀等预处理提取出车牌字符;然后收集字符数据,对CNN神经网络在角度倾斜、光照变化和噪声污染复杂条件下进行训练,使用训练后得到的模型对车牌字符图片进行识别.实验结果表明,该方法识别车牌字符正确率较高,在角度倾斜、光照变化等噪声污染条件下的准确性和稳定性较好,能够有效地降低车标识别的错误率.In view of inefficiency and poor reliability of traditional license plate character detection methods,a convolution neural network license plate character recognition method based on Haar cascade detection conbined with deep learning is proposed.Haar cascade classifier is used to detect the position of license plate in the picture,and then the license plate character is extracted by pretreatment technology such as grayscale,threshold value,corrosion and expansion.Through the collection of character data,training the CNN neural network under the complicated conditions of angel tilt,light change and noise pollution,then using the model obtained after training to recognize the license plate character picture.The experimental results show that the method can recognize the license plate characters effectively,and it has good accuracy and stability in light change and noise pollution thus reducing the error rate of license plate recognition effectively.

关 键 词：车牌识别 图像处理 神经网络 深度学习 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]