基于支持向量机的纸币识别方法设计与研究  

Design and Research of Paper Currency Identification Based on Support Vector Machines

作　　者：蒋琳琼[1] 贺建飚[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《计算机测量与控制》2007年第7期937-939,958,共4页Computer Measurement &Control

基　　金：国家自然科学基金资助项目(60503007);国家"八六三"计划资助项目(2003AA1Z2190);国家"十五"科技攻关资助项目(2003BA104C)

摘　　要：纸币识别是一个小样本、非线性和高维模式识别问题,是当前模式识别中的难题之一,具有重要研究意义和实用价值;选用支持向量机二次优化算法中的序贯最小优化算法,该算法以解析的方法处理优化问题,训练速度较快,识别率较高;序贯最小优化算法优化标准的单一阈值容易错判优化条件,从而导致花费大量时间寻找第二个优化样本;在优化标准中增设上下界两个阈值来判断优化条件,避免了原算法单一阈值判决的这个缺点,加快了训练速度,提高了识别率;将此种支持向量机训练算法用于纸币识别,能够充分发挥支持向量机解决小样本、非线性和高维模式识别问题的优点,能够适合工程应用中的需要。Paper currency identification, a scared samples, nonlinear and high dimensions pattern recognition problem is one of the difficult problems of modern pattern recognition and of specific research significance and practical value. This paper selects sequential minimal optimization algorithm of the quadratic optimization algorithms of support vector machine. The sequential minimal optimization algorithm deals with the optimization problem by explicit method, with high training rate and high identification rate. The sequential minimal optimi- zation algorithm optimizes the traditional single threshold error tolerance optimization condition, which results in the time-consume to seek the second optimization sample. Two thresholds, upper and lower are added to judge the optimization conditions. This avoided the original disadvantage, accelerated the training rate and improved identification rates. This training algorithm of support vector machine has been used in paper currency identification, shows the advantages of capability in dealing with scared samples, nonlinear and high dimensionsand can meet the oroiect demands.

关 键 词：支持向量机 纸币识别 二次优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]