基于SVM分类器集装箱箱号识别法.docVIP

基于SVM分类器集装箱箱号识别法 　　摘 要：为准确、高效地识别集装箱箱号，提出基于支持向量机（Support Vector Machine, SVM）分类器的集装箱箱号识别法.在对大量箱号图片进行实验并统计各种特征识别率的基础上，经过预处理、箱号定位、字符分割，得到36×22像素大小的二值化图像；提取箱号字符的边界和质心特征、改进的灰度直方图特征以及边缘方向直方图特征，将这些特征进行组合形成特征向量并进行降维和归一化处理；利用SVM分类器把处理过的特征向量进行分类并识别.实验结果表明，该方法平均识别正确率可以达到95%，高于使用单一特征的识别、简单的模板匹配算法以及特征加权（八邻域）模板匹配算法. 　　关键词：字符识别； 支持向量机； 质心特征； 改进的灰度直方图特征； 边缘方向直方图特征 　　中图分类号：U691.1； TP391.41文献标志码：A 　　 　　Recognition method of container code based on SVM classifier 　　AN Bowen, LI Dan, PANG Ran 　　(Information Engineering College, Shanghai Maritime Univ., Shanghai 201306, China) 　　Abstract： In order to recognize container code precisely and effectively, a recognition method of container code based on support vector machine (SVM) classifier is proposed. On the basis of experiment by using many container code images and count recognition rate of various features, binary images with size of 36×22 pixels are obtained by preprocessing, container code locating and character segmentation. The character features including edge, center of mass, improved grayscale histogram and edge direction histogram are extracted and combined into feature vector which is processed by dimensionality reduction and normalization. The processed feature vector is classified and recognized through SVM classifier. Experiment results show that the average recognition rate can reach 95%, which is higher than that of single feature recognition, simple template matching algorithm and weighted feature (8-neighborhood) template matching algorithm. 　　Key words： character recognition; support vector machine; center of mass feature; improved grayscale histogram feature; edge direction histogram feature 　　 　　0 引 言 　　在集装箱码头、海关关口、货场等场所进出的集装箱都需要进行检测.每个集装箱上均标有唯一的11位箱号代码，它们由箱主代码（3位英文字母）、设备识别码（字符“U”）、箱体注册码（6位数字）和校验码（1位数字）组成［1-2］，如何准确、高效地识别11位箱号代码是实现集装箱智能道口的关键技术.在准确定位、分割箱号字符后，一般采用简单的模版匹配算法、特征加权（八邻域）模板匹配算法以及神经网络算法等进行箱号识别.［3-5］但在实际应用中，这3种算法往往由于识别精度差或计算速度慢而遭弃用. 　　支持向量机（Support Vector Machine，SVM）是机器学习界的研究热点，被成功地应用于分类、函数逼近和时间