卷积神经网络在车牌分类器中的应用

发布时间：2018-02-09 09:32

本文关键词： 车牌分类卷积神经网络卷积层降采样层　出处：《计算机工程与应用》2017年14期 　论文类型：期刊论文

【摘要】：传统的基于边缘、颜色、纹理及机器学习等方法进行的车牌定位,需要对车牌图像进行复杂的特征提取,不但训练过程容易造成过拟合或者维数灾难,而且识别结果也易受光照、道路环境及图像质量等因素的影响,虽然漏识别率低,但误识别率高。针对车牌分类问题,利用深度学习中的卷积神经网络,避免了传统模式分类算法在前期对图像复杂的预处理,降低了设计提取特征算法时对丰富经验的依赖。综合对比了BP神经网络、支持向量机、卷积神经网络三种算法,实验结果表明,卷积神经网络在车牌分类中具有较好的表现,识别率高达98.25%,也证明了深度学习在智能交通领域具有较大的应用前景。
[Abstract]:Traditional license plate location based on edge, color, texture and machine learning requires complex feature extraction of license plate image, not only the training process is easy to cause over-fitting or dimension disaster. The recognition results are also easily affected by illumination, road environment and image quality. Although the rate of missing recognition is low, the rate of false recognition is high. In view of the problem of license plate classification, the convolution neural network in depth learning is used. It avoids the complex preprocessing of image by traditional pattern classification algorithm and reduces the dependence on rich experience in designing feature extraction algorithm. Three algorithms, BP neural network, support vector machine and convolutional neural network, are compared synthetically. The experimental results show that the convolution neural network has a good performance in license plate classification, and the recognition rate is as high as 98.25. It also proves that the deep learning has a great application prospect in the field of intelligent transportation.
【作者单位】：北京工商大学材料与机械工程学院;
【基金】：交通运输部信息化科技项目(No.2012-364-835-110) 北京工商大学2017年研究生科研能力提升计划项目
【分类号】：TP183;TP391.41

【相似文献】