基于向量机学习算法的多模式分类器的研究及改进  

Research and improvement of multiclass pattern classifier based on vector machine learning algorithm

作　　者：柳长源[1,2] 毕晓君[1] 韦琦[2] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]哈尔滨理工大学电气与电子工程学院,黑龙江哈尔滨150080

出　　处：《电机与控制学报》2013年第1期114-118,共5页Electric Machines and Control

基　　金：国家自然科学基金(61175126);中央高校基本科研业务费专项资金(HEUCFZ1209);教育部博士点基金(20112304110009);黑龙江省自然科学基金(E200932)

摘　　要：为了提高向量机"一对一"学习算法在多模式识别中的分类效率,对基于支持向量机和相关向量机算法进行多模式分类的方法进行研究,发现比较次数过多是该方法计算量大的主要原因。提出了一种在每轮比较中,排除最差类别的新方法。该方法使比较次数逐级减少,并且当类别数较多时,总计算量减少尤其明显。通过理论分析和对数据分类的实验结果表明,新方法与传统分类器相比,在基本不影响分类正确率的前提下,机器训练与识别次数显著减少,算法运行速度明显提高。In order to improve classification efficiency of multiclass pattern recognition based on ＂one a- gainst one＂ learning algorithm in vector machine, investigated the method of support vector machine and relevance vector machine algorithm in multi-mode classification, and found that comparison for too many times was the main reason for large amount of calculation. Proposed a way that eliminated the most dis- similar class in each round of comparison. Comparison times were reduced step by step per cycle. The classification number was more, and the decrease in the total calculation amount was more obvious. The theory analysis and the experimental results of data classification show that compared with traditional clas- sifier, the training times and the recognition times of the method are greatly reduced under the premise of hardly influencing classification accuracy, and the algorithm running speed is improved obviously.

关 键 词：模式识别 支持向量机 相关向量机 分类器 “一对一”算法 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]