基于支持矢量机特征空间距离的多目标分类方法  

A Multi-Target Classification Method Based on Distance of Support Vector Machine Feature Space

作　　者：郭雷[1] 肖怀铁[1] 付强[1] 

机构地区：[1]国防科技大学电子科学与工程学院ATR实验室,湖南长沙410073

出　　处：《计算机仿真》2005年第9期272-274,共3页Computer Simulation

摘　　要：支持矢量机是近年来在统计学习理论的基础上发展起来的一种新的模式识别方法,主要解决的是两类目标分类问题。多类目标分类一般是分解为多个两类目标分类。多类目标分类之后通常会对某些目标错误分类或者某些目标不能判定其类别,这就是支持矢量机多类目标分类中的错分、拒分现象。针对这个问题,该文提出了一种基于支持矢量机特征空间距离的模糊隶属度函数,根据模糊隶属度的大小对错分和拒分目标重新分类。对美国资源探测卫星数据的多目标分类仿真结果表明,采用这种方法重新分类后,能够有效地减少错分和拒分目标的数量,提高了正确识别率。Support vector machine is a new learning method developed in recent years based on the foundation of statistical learning theory, and mainly solves two -class targets classification. The problem of multi -target classification is usually solved by a decomposing and reconstruction procedure of two - class decision machine. Some targets may be misclassified or could not deternline which class they belong to after multi - target classificaton, this is called misclassification and rejected classification. To resolve this problem, we propose a new fuzzy membership function based on the distance of support vector machine feature space, and then reclassify the misclassified and reject - classified targets according to the fuzzy membership. Simulation results of Landsat MSS imagery multi - target recognition show that it can effectively reduce the number of misclassification and rejected classification targets by using this method, and also can improve correct recognition ratio consequently.

关 键 词：支持矢量机 多目标识别 特征空间 模糊隶属度函数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]