【作者】 陆璐；

【导师】 高隽； 张旭东；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2006， 硕士

【摘要】 人工神经网络是一种模拟人脑功能并具有广泛应用前景的网络,卷积神经网络是其中一种,它具有权值简单、全局优化、适应性强、理论完备、泛化性能好等优点,神经网络及其特例卷积神经网络是目前机器学习领域的研究热点。随着通信、信息和电子技术及计算机网络技术的发展,车牌识别系统及其应用正受到越来越多国家的重视。 本文在研究卷积神经网络的基础上将其应用到车牌识别中,主要进行的工作如下: (1)整理总结了国内外学术界关于神经网络方面的研究成果,介绍神经网络的基本概念和基本原理; (2)卷积神经网络作为神经网络中的一种特殊网络,本文在国内外学术界对其研究成果的基础上,详细阐述了卷积神经网络的基本原理,分析其基本结构和网络参数; (3)在对卷积神经网络分析研究的基础上,先将卷积神经网络应用于形状识别任务中,证明网络高效的识别能力,再将改进后的卷积神经网络应用于车牌识别问题中:先利用牌照区域灰度变化迅速的特点定位车牌,然后采用修正投影方法分割车牌字符,在最后识别阶段采用卷积神经网络进行字符识别,并与其他传统学习方法进行了对比取得了较为满意的结果。更多还原

【Abstract】 Artificial Neural Network(ANN) is a promising network. It simulates human being’s cerebrum. Convolutional Neural Network(CNN) is one kind of them, which has the advantages of global solutions, good adaptability, high generalization ability and maturity in theory. Both ANN and CNN are hot spots in the field of machine learning nowadays. With the development of communication, information and electronic technology and computer network, License Plate Recognition System(LPR) attracts much more attentions.In this paper, we introduce CNN to the field of LPR. The main work is described as follows:(1) We summarize the latest research achievements and development of ANN, present the conceptions and the principles of it;(2) As one of ANN, CNN is studied in detail. We summarize the development and latest research achievements of it. Furthermore, we expatiate on its basic principles, analyze its structure and parameters.(3) Then we use CNN to settle the certain tasks. Firstly, we use it to solve the shape recognition problem. In the stage of test, the generalization ability of CNN is better than other methods. Secondly, CNN is used in the problem of license plate recognition. Because the gray of the region of license changes rapidly, we make use of this characteristic to locate the license first, and then vertical projection is used to segment the characters in this paper, in the stage of character recognition we adopt CNN method and the result is satisfied.更多还原