卷积神经网络在路牌识别中的应用  

Application of Convolutional Neural Network in Identification of Guideboards

作　　者：宋波[1] 王启春[2] 

机构地区：[1]四川省交通运输厅公路规划勘察设计研究院,成都610000 [2]重庆工程职业技术学院,重庆402260

出　　处：《公路交通技术》2015年第5期114-118,125,共6页Technology of Highway and Transport

基　　金：国家自然科学基金青年基金项目(41104011)

摘　　要：卷积神经网络是近年来广泛应用于模式识别、图像处理等领域的一种高效识别算法,其具有结构简单、训练参数少和适应性强等特点。简要介绍卷积神经网络(CNN)的发展现状和基本原理,并基于CNN设计提出一种路牌识别方法:首先,对数据进行预处理;其次,建立CNN并设置训练参数,对其进行训练;最后,利用测试样本对训练结果进行验证。验证结果表明:利用该方法来识别道路限速牌可以达到较高的准确率。The Convolutional Neural Network is an efficient identification algorithm that has been extensively applied in fields of mode identification and image processing, etc. in recent years and is characterized by simple structure, few training parameters and strong adaptability, etc. This paper briefly introduces the development status and basic principle of the Convolutional Neural Network （ CNN ） , and based on CNN design proposes a guideboard identification method： first, preprocess data; second, establish CNN and set training parameters; and finally validate the training results by means of test samples. The results of validation show that identification of speed restriction boards on road by means of this method can achieve higher accuracy.

关 键 词：卷积神经网络(CNN) 路牌识别 机器学习 

分 类 号：U491.52[交通运输工程—交通运输规划与管理]