基于卷积神经网络的车牌识别技术研究

发布时间：2018-03-31 16:09

本文选题：卷积神经网络　切入点：智能交通系统　出处：《电子科技大学》2017年硕士论文

【摘要】：卷积神经网络是可以模拟人的大脑功能并能够应用到更多领域的一种特殊神经网络,它是近年发展起来,相比于其他同类方法,卷积神经网络具有理论完备、泛化性能好、全局性能优化、适应性强等优点。卷积神经网络是目前机器学习领域的研究热点。作为一种新兴技术,卷积神经网络在很多应用领域的研究还不成熟,有待进一步的探索和完善。现如今,利用高新技术,智能交通系统对传统的交通系统进行的改造,发挥着巨大的效能也获得了深厚的社会经济效益。随着计算机网络技术和通信技术的逐步发展,车辆牌照识别系统在越来越多的国家扮演着举足轻重的角色。在现实生活中传统的车牌识别系统,早期阶段的预处理可能导致车牌字符分割和定位不清的错误和缺点,这将影响到车牌识别的效果,减少实际的识别率。并且,传统车牌识别方法的图像预处理过程耗时,无法应对实际应用中的实时性要求,并且容易受到噪声影响,难以充分保留原始信号,会进一步降低识别效果。本文在研究分析卷积神经网络工作机理的基础之上,将局部权值共享的卷积神经网络方法引入智能交通系统这一具体的应用领域,通过多维网络输入向量图像可以直接输入这一特性,能够在图像识别和处理方面有较好的效果,避免了在特征提取的过程中的复杂度。本文针对基于卷积神经网络下的车辆牌照识别研究课题,整理归纳了国内外学术界的研究现状和成果,介绍了利用卷积神经网络进行图像识别的原理。在对经典神经网络结构LeNet-5的分析研究基础上加以完善,将完善后的卷积神经网络ILeNeT-5应用于车牌识别问题中,并基于MATLAB平台,完成应用程序的开发,最终完成基于卷积神经网络下的车辆牌照识别的研究工作。本文所研究的基于卷积神经网络下的车牌识别,是在神经网络的优势下,使用一种改进的ILeNeT-5神经网络对车牌的识别,它优化了网络中卷积层和采样层的参数,在特殊情境下也提高了车牌的识别率,能有效的提高车牌识别度,对于智能交通系统的建设具有重大的社会意义。
[Abstract]:Convolutional neural network is a special neural network which can simulate human brain function and can be applied to more fields. It has been developed in recent years. Compared with other similar methods, convolutional neural network has perfect theory and good generalization performance. Global performance optimization, strong adaptability and so on. Convolution neural network is the research hotspot in the field of machine learning. As a new technology, the research of convolution neural network in many application fields is not mature. Need to be further explored and improved. Nowadays, the transformation of traditional transportation systems by using high and new technologies and intelligent transportation systems, With the development of computer network technology and communication technology, Vehicle license plate recognition system plays an important role in more and more countries. In the traditional license plate recognition system in real life, the preprocessing of early stage may lead to the errors and shortcomings of license plate character segmentation and unclear location. This will affect the effect of license plate recognition and reduce the actual recognition rate. Moreover, the image preprocessing process of the traditional license plate recognition method is time-consuming, unable to meet the real-time requirements of practical applications, and vulnerable to the impact of noise. It is difficult to fully retain the original signal, which will further reduce the recognition effect. In this paper, based on the analysis of the working mechanism of convolution neural network, The convolution neural network method of local weight sharing is introduced into the specific application field of intelligent transportation system. The multi-dimensional network input vector image can directly input this characteristic, and it has good effect in image recognition and processing. The complexity of feature extraction is avoided. In this paper, the current research situation and achievements of the domestic and foreign academic circles are summarized for the vehicle license plate recognition based on convolutional neural network. This paper introduces the principle of image recognition using convolutional neural network. Based on the analysis and research of classical neural network structure LeNet-5, the improved convolutional neural network ILeNeT-5 is applied to license plate recognition, and based on MATLAB platform. The research work of vehicle license plate recognition based on convolution neural network is finished, and the license plate recognition based on convolution neural network is studied in this paper, which is based on the advantage of neural network. An improved ILeNeT-5 neural network is used to recognize license plate. It optimizes the parameters of convolution layer and sampling layer in the network, and improves the recognition rate of license plate under special circumstances, which can effectively improve the recognition degree of license plate. It is of great social significance to the construction of intelligent transportation system.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【参考文献】