摘要： 提出一种端元约束条件下的非负矩阵分解方法来自动反演混合像元组分。以端元光谱之间的差距为约束条件，使得目标函数综合了影像的分解误差和端元光谱的影响，并以最大后验概率方法导出了限制性非负矩阵分解的迭代算法。成像光谱数据实验结果表明该方法能够自动提取影像的端元光谱矩阵与组分信息，且分解精度比IEA方法高。

关键词: 非负矩阵分解, 混合像元, 约束, 高光谱

Abstract: An endmember constrained Nonnegative Matrix Factorization(NMF) method for mixture pixels unmixing is proposed. A penalty function imposed by maximizing difference between endmembers among all possible simplexes is presented in the image. The maximum posterior probabilistic method is utilized to formulate a novel iterative algorithm by trading off abundance nonnegative attribution and endmember difference. Experimental result with PHI data shows the proposed algorithm is an alternative approach for endmember abstraction and abundance estimation. Comparison with IEA validates the efficience of the proposed method.

Key words: Nonnegative Matrix Factorization(NMF), mixture pixels, constrainted, hyperspectral imagery

中图分类号: 

• TP237.3