基于DS聚类的高光谱图像集成分类算法

Hyperspectral image ensemble classification algorithm based on dominant set clustering

针对高光谱遥感图像维数高、样本少导致分类精度低的问题,提出一种基于DS聚类的高光谱图像集成分类算法(DSCEA)。首先,根据高光谱数据特点,从整体波段中随机选择一定数量的波段,构成不同的训练样本;其次,分析图像的空谱信息,构造无向加权图,利用优势集(DS)聚类方法得到最大特征差异的波段子集;最后,根据不同样本,利用支持向量机训练具有差异的单个分类器,采用多数表决法集成最终分类器,实现对高光谱遥感图像的分类。在Indian Pines数据集上DSCEA算法的分类精度最高可达到84.61%,在Pavia University数据集上最高可达到91.89%,实验结果表明DSCEA算法可以有效地解决高光谱图像分类问题。

To solve the low classification accuracy problem of hyperspectral image classification due to the high dimensionality and the low sample size,this paper developed a hyperspectral image classification ensemble algorithm based on DS clustering.Firstly,according to the characteristic of hyperspectral data,it selected a certain number of bands from the whole bands randomly,to constitute different training samples.Secondly,it constructed a weighted undirected graph by studying the spectral and spatial information of hyperspectral image,and obtained the maximum characteristic difference of band set by utilizing DS clustering approach.Finally,according to the different samples,it trained a single classifier with different differences by SVM,obtained the ensemble classifier by the majority voting method,to realize the classification problem of hyperspectral image.The experimental results show that classification accuracy of DSCEA can reach 84.61%on the Indian Pines dataset,and the classification accuracy of DSCEA can reach 91.89%on the Pavia University dataset,it validated the DSCEA algorithm can solve the hyperspectral image classification problem.