摘要
在进行高光谱混合像元非线性分解应用中,提出一种非监督的高光谱混合像元非线性分解方法.通过核函数把原始高光谱数据映射到高维特征空间中,揭示数据之间的高阶性质.通过非线性映射,原始数据在高维特征空间中变得线性可分.在高维特征空间中运用线性的非负矩阵分解(NMF)算法进行光谱解混,挖掘出数据间更多的特征.解混结果以端元相关系数、光谱角距离、光谱信息散度和均方根误差作为质量评价指标.进行模拟数据仿真实验和真实高光谱遥感数据分解实验,结果表明,采用该算法得到的分解结果优于非负矩阵分解算法.
An unsupervised nonlinear decomposing algorithm for hyperspectral imagery was introduced to solve the nonlinear decomposing problem of hyperspectral imagery. The original data were mapped into a high-dimensional feature space by a nonlinear mapping, which was associated with a kernel function. Then the higher order relationships between the data were exploited. The mapped data became linearly separable in the high-dimensional feature space by using an appropriate nonlinear mapping. Then a linear nonnegative matrix factorization （NMF） method can be applied to extract more useful features. gndmember correlation coefficient, spectral angle distance, spectral information divergence and root mean square error were used to estimate the quality of the results. The experimental results of synthetic mixtures and a real image scene demonstrated that the method outperformed the nonnegative matrix factorization approach.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2011年第4期607-613,共7页
Journal of Zhejiang University：Engineering Science
基金
浙江省自然科学基金资助项目(Y1100196)
关键词
混合像元
核函数
光谱分解
非负矩阵分解(NMF)
mixed pix
kernel function
spectral decomposing
nonnegative matrix factorization （NMF）
作者简介
厉小润（1970-），男，浙江东阳人，副教授，从事模式识别和遥感图像分析的研究．E-mail：lxr@zju．edu．cn
