先验信息约束NMF的高光谱解混

Hyperspectral unmixing of nonnegative matrix factorization with prior information

针对高光谱图像中含有大量混合像元,且大多数解混算法未能利用真实地物信息的问题,提出了一种利用先验信息约束的非负矩阵分解方法对高光谱进行解混。首先利用顶点成分分析法和全约束最小二乘法分别对端元矩阵和丰度矩阵进行初始化,然后利用本文算法对高光谱数据进行解混,最后对估计端元和估计丰度进行评价分析。实验显示,利用本文提出的方法对数据解混的结果优于其他约束的非负矩阵分解算法得到的结果,在求解过程中有很好的抗噪性能。

In view of the problem that hyperspectral images contain a large number of mixed pixels,and most of the de?mixing algorithms fail to utilize real object information,this paper proposes a non?negative matrix factorization method using a priori information constraint.Firstly,the end?element matrix and the abundance matrix were initial?ized respectively by the vertex component analysis method and the full constrained least squares method.Then the hyperspectral data was de?mixed by the algorithm,and finally the estimated end?member and estimated abundance were evaluated.Experiments show that the results obtained by the method proposed in this paper are better than those obtained by non?negative matrix decomposition algorithms of other constraints,and have good anti?noise per?formance in the solution process.