高光谱影像的BDT-SVM地物分类算法与应用

面对海量数据的特征空间高维性及训练样本的有限性,高光谱遥感影像若采用常规统计模式的分类方法难以获得较好的分类结果。因此探讨支持向量机(SVM)分类器的基本原理,针对EO-1Hyperion高光谱影像的分类特点及现有多类SVM算法所存在的训练时间长及分类精度低等问题,引入二叉决策树SVM(BDT-SVM)分类算法,并提出一种新的类间分离度定义方法及相应的客观确定二叉树结构的策略,由此生成改进的BDT-SVM算法。实验结果表明:与其他多类分类方法相比,基于改进的BDT-SVM算法的高光谱影像地物分类效果更好,总体精度达到90.96%,Kappa系数为0.89,该算法还解决了经典SVM多类分类可能存在的不可分区域问题。

The Object Classification Algorithm and Application for Hyperspectral Imagery based on BDT-SVM

Hyperspectral remote sensing is a cutting edge field in remote sensing.It offers the fine detection of objects by its spectral response characteristics in various spectral bands,and has superiority to multispectral remote sensing in fine extraction.However,due to high\|dimensional feature space and limited training samples of the huge data of hyperspectral images,it is difficult for conventional statistical pattern identification methods to classify hyperspectral images.Thus this paper explores the basic principle of support vector machine classifier and employs Binary Decision Tree Support Vector Machine (BDT\|SVM) classification algorithm based on EO\|1 Hyperion hyperspectral imagery.And this study proposes a new definition of the class separation against the long training time and low classification efficiency of existing multi\|class SVM algorithm and generates a modified BDT\|SVM algorithm.On the basis of theoretical analysis,this paper completes the object classification experiments on Hyperion hyperspectral imagery of the test area and verifies the high classification accuracy of the method.Experimental results show that the effect of hyperspectral image classification based on the modified BDT\|SVM algorithm is apparently better than other multi\|class classification methods,which total classification accuracy is up to 90.96% and kappa coefficient is 0.89.The algorithm also solves the problem of non\|separable region,which may be present in the classic SVM multi\|class classification methods.