基于ETAFSVM的高光谱遥感图像自动波段选择和分类  

Automatic Band Selection and Classification of Hyperspectral Remote Sensing Images Based on ETAFSVM

作　　者：戴宏亮[1,2] 戴道清[1] 

机构地区：[1]中山大学数学与计算科学学院,广州510275 [2]广东商学院数学与计算科学学院,广州510320

出　　处：《计算机科学》2009年第4期268-272,共5页Computer Science

基　　金：国家自然科学基金(NSFC#60575004,NSFC#10771220);教育部高等学校博士点科研基金(SRFDP-20070558043)资助

摘　　要：提出了一种新型的具有良好特性的支持向量机——全间隔自适应模糊支持向量机(TAFSVM),并提出一种新的遗传算法——智能遗传算法(IGA)来设计一个TAFSVM分类器,称为ETAFSVM,同时优化高光谱遥感图像自动波段选择和TAFSVM参数集,并且结合5-fold交叉验证来确定其泛化能力,最后将ETAFSVM应用于高光谱遥感图像数据。通过先进行自适应波段选择后再用径向基神经网络分类器、K-最近邻分类器和标准支持向量机等3种方法进行全部分类精度比较,以及与这3种方法直接进行类别分类精度和平均分类精度比较,其结果表明运用ETAFS-VM不仅可以自动进行波段选择,而且分类精度较高,对Hughes现象敏感性较低,是进行高光谱遥感图像分类的一种有效方法。In this study,total margin-based adaptive fuzzy support vector machine（TAFSVM） which has good quality was proposed. Besides, this paper proposed an evolutionary approach to design a TAFSVM-based classifier（named ETAFSVM） by simultaneous optimization of automatic band selection and parameters tuning using an intelligent genetic algorithm（IGA）, combined with 5-fold crossvalidation regarded as an estimator of generalization ability. Subsequently, the model of ETAFSVM was used to classify hyperspectral remote sensing images. Comparing with adaptive bands selecting firstly, then using radial basis functions neural network, K-nearest neighbors classifier and standard SVM to classify the test data for overall classification percentage accuracy, and then using the three classifiers to classify the test data for class and average percentage accuracy. The experimental results indicate that the proposed ETAFSVM model can achieve both higher classification accuracy and lower sensitivity to the Hughes phenomenon. Consequently, the ETAFSVM model provides a promising alternative for classification in hyspectral remote sensing images.

关 键 词：全间隔自适应模糊支持向量机 智能遗传算法 高光谱遥感图像 分类 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]