基于监督等距映射高光谱遥感影像降维

摘要

Considering the intrinsic nonlinear structure of hyperspectral remote sensing data and the characteristic of unsupervision of traditional manifold learning, during the process of dimension reduction of classification-oriented hyperspectral remote sensing data, we propose a new method of supervised isometric mapping (S-Isonmap). The method is based on the idea that the between-class distance is greater than the within-class distance. First it obtains initial category labels of the samples by using KMEANS algorithm on primary data for clustering; then it searches the K-Nearest neighbour of the data points with new distances, and further executes the dimension reduction by Isomap. Experiments demonstrate that the presented method outperforms the traditional Isomap.%在面向分类的高光谱遥感数据降维过程中,考虑到高光谱遥感数据内在的非线性结构和传统流形学习非监督的特点,提出一种新的监督等距映射方法(S-Isomap).方法基于类间距离大于类内距离的思想,首先利用KMEANS算法对原始数据进行聚类得到样本的初始类别标签,采用新距离搜寻数据点的K近邻,进而实施等距映射降维.实验证明了该方法优于传统Isomap.