APP下载

基于不同降维方法的PPI端元提取效果对比研究

2017-02-15黄晨张运张立伟

安徽农学通报 2017年1期

黄晨+张运+张立伟

摘 要:传统PPI算法采用最大噪声分离(MNF)方法进行降维,MNF变换中均设定数据之间线性相关,在某些情况下会使变换后的结果具有某些人为特征,在降维过程中会丢失信号较弱的信息,导致端元数量少;分段主成分分析(SPCA)降维方法具有不改变图像的物理意义,且信息保存较完整的优势。该研究采用不同降维方法利用纯净像元指数法(PPI)对不同下垫面地表提取端元,结果表明,在地表破碎区域SPCA降维后可找出信号较弱的端元提取的端元数量多与MNF降维提取的端元数,而地物聚集区MNF降维方法提取的端元质量更好。研究结果可以为不同下垫面的高光谱影像端元提取以及降维方法的选择提供参考。

关键词:端元提取;最大噪声分离;分段主成分分析;纯净像元指数法

中图分类号 TH744.1 文献标识码 A 文章编号 1007-7731(2017)01-0013-06

Abstract:The traditional PPI algorithm uses the maximum noise separation (MNF) to reduce dimension,MNF transform was set linear correlation between the datas,and in some cases,the results of the transform will have some human characteristics.In the process of dimension reduction,the weak signal will be lost,which leads to a small number of end elements.Segmented principal component analysis (SPCA) does not change the physical meaning of the image and the information will be relatively complete preservation.This paper use the pure pixel index (PPI) for different dimensionality reduction methods and for different underlying surface to extract endmember.The results shows that the SPCA dimensionality reduction is more suit in broken underlying surface,it could find the weak signal endmember;and the MNF dimensionality reduction will find the better quality endmember in ground gathering area.The results of this research can provide reference for the endmember extraction of hyperspectral image and the selection of dimension reduction method for different underlying surfaces.

Key words:Endmember extraction;MNF;SPCA;PPI

1 引言

混合像元分解是高光谱数据处理中较为重要的组成部分,混合像元的分解分为两步,第一步是端元提取,即某一像元对应区域内仅存在一种地物类型,这个地物的光谱就是要提取的端元;第二步是丰度反演,即不同纯净地物所占像元的比例,端元提取是混合像元分解的重要前提和关键步骤。纯净像元指数法(Pure Pixel Index,简称 PPI)[1-2]是以线性光谱混合模型的几何学描述为基础,利用端元是遥感图像在特征空间中所形成的单形体的端点的特点、单形体的向量投影性质进行端元提取。PPI算法在端元提取中较为成熟,且便于实现,方法灵活,很多软件都将此方法作为端元提取的模块。……

登录APP查看全文