摘要
支持向量机(SVM)是一种崭新的机器学习方法,建立在结构风险最小化原理基础上,寻找一个最优分类超平面,引进核函数将低维空间向量映射到高维空间.此方法能解决小样本、非线性及高维模式识别中的问题.鉴于此,将SVM应用于多传感器信息融合,并针对多类型目标识别问题,采用“oneagainstall”方法构造多元分类器.实验中比较了采用不同核函数构造的SVM的分类效果,结果表明SVM具有较高的识别率,其中三项多项式核函数构造的SVM的识别率最高,可达到93.2%.另外,还比较了单传感器和多传感器融合的识别结果,单传感器的识别率只有63.7%,大大低于多传感器融合的识别率.
Support Vector Machine (SVM) is based on Structure Risk Minimization principle (SRM) is a kind of machine learning method. Many problems with small samples, nonlinearity or high dimension in pattern recognition could be solved by this machine. In this paper, SVM was applied to research multi-sensor data fusion. For multi-target recognition, an approach of “one-against-all” was used to construct multiple class binary classifier. In experiments, the classification effects were based on SVM with different kernel functions were compared. The results showed the high efficiency of recognition and classification by SVM. Especially, SVM with three-polynomial kernel function has the highest classification effect by 93.2%. In addition, the effects with single sensor and multiple sensors were also compared. The recognition rate with single sensor is only 63.7%, being greatly lower than that with multiple sensors.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2005年第2期41-43,共3页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国防预研基金资助项目(00J16.6.3.JW0401).
关键词
支持向量机
数据融合
目标识别
support vector machine
data fusion
target recognition
