基于主成分判别分析的高光谱遥感影像分类方法

Hyperspectral Remote Sensing Images Classification Method Based on Principal Component Discriminant Analysis

提出了一种基于主成分判别分析的高光谱遥感影像分类方法。针对高光谱遥感影像数据量大、冗余信息多的特点,使用改进的线性判别分析方法对高光谱遥感数据进行线性维数减少。该方法将主成分分析加入到线性判别分析的算法框架中,能够克服常规的线性判别分析方法在训练样本数量较少时遭遇到的小样本问题。通过实验,证明基于主成分判别分析的遥感影像分类方法能够利用少量的训练样本实现更优的分类精度。

In this paper, we proposed a hyperspectral remote sensing images features extraction and classification method based on principal component discriminant analysis. In order to overcome the high dimensionality and high redundancy of hyperspectral images, we used modified linear discriminant analysis to reduce the linear dimension of hyperspectral data. In particular, the principal component analysis(PCA) was introduced into the conventional linear discriminant analysis(LDA) framework to deal with the small sample size(SSS) obstacle caused by limited training samples in learning problem. Experiments on hyperspectral dataset demonstrate that the proposed method can deliver high classification rate with a small number of training samples.