摘要
基模分析被广泛地运用于机器学习与数据挖掘之中,其核心思想是通过寻找数据凸体的角点,这些角点通常是数据的主要成分。同时通过分配比例系数给这些角点,这样就能重构原始数据。本文据此出发,利用基模分析进行高光谱盲信号分离,分别分离出端元矩阵与各种物质的比例系数。在梯度下降算法的框架下,我们采用了一种快速初始化策略,利用基模分析的扩展模式-核方法进行端元与比例系数的迭代。通过对真实高光谱遥感影像大量实验发现,此方法简单易行,且精度较高。
The archetypal analysis is widely used in machine learning and data mining, and its core is to find these comers of data convex, and these comers are usually the main components of datas. If we update the coefficient matrix of datas,and the archetypal analysis become the blind signal separation algorithm.In the paper,we use the archetypal analysis as blind signal separation tool,and separate the original hyperspectral image datas into endmembers and the proportion coefficients matrixes of various substances. In the framework of gradient descent algorithm,we use a fast initialization strategy-FUR_THESTSUM algorithm,and use the extended model of archetypal analysis-kernel algorithm to update the two matrixes.A great numbers of experiments on real hyperspectral image show that archetypal analysis is an efficient blind signal separation,and the algorithm is have high accuracy.
出处
《中国建材科技》
2014年第2期72-76,共5页
China Building Materials Science & Technology
关键词
高光谱混合像元分解
盲信号分离
梯度下降算法
基模分析
hyperspectral image mixed pixels unmixing
blind signal separation
gradient descent algorithm
archetypal analysis
