基于SVM不同核函数的多源遥感影像分类研究  

Research on multi-source remote sensing image classification based on SVM different kernel functions

作　　者：王双亭[1] 艾泽天[1] 都伟冰[1] 康敏[1] 

机构地区：[1]河南理工大学测绘与国土信息学院,河南焦作454000

出　　处：《河南理工大学学报（自然科学版）》2011年第3期304-309,共6页Journal of Henan Polytechnic University（Natural Science）

基　　金：国家重点基础研究发展计划项目(2009CB226100)

摘　　要：选取同地区同时相的多光谱和高光谱影像,在实验样本和验证样本相同的情况下,采用SVM分类算法中4种不同的核函数,对2种影像进行分类实验.结果表明,对于多光谱影像,RBF核函数分类精度最高,Sigmoid最低;对于高光谱影像,Linear核函数分类精度最高,Sigmoid最低;对于同地区相同分辨率的遥感图像,在分类条件相同的情况下,多光谱影像的分类精度和高光谱的分类精度相近.The paper uses multi-spectral image and hyperspectral image of the same time in the same area as the research target,and employs four different kernel functions of SVM classification algorithm to make experiments between these two images based on the premise that the research has the same test samples and identifying samples.The experiments show that for multi-spectral image,RBF kernel function classification will produce the maximum classification precision,while Sigmoid is at its minimum;for hyperspectral images,Linear kernel function will achieve the maximum classification precision,and Sigmoid is at its minimum;for the same resolution remote sensing images at the same area,on the condition of the same classification standard,the classification precision of multi-spectral image is similar to that of hyperspectral image.

关 键 词：SVM 核函数 多源遥感影像分类 

分 类 号：TP394[自动化与计算机技术—计算机应用技术]