基于再权重稀疏和正交约束非负矩阵分解的高光谱图像解混

REWEIGHT SPARSE AND ORTHOGONAL NONNEGATIVE MATRIX FACTORIZATION FOR HYPERSPECTRAL UNMIXING

针对由于非负矩阵分解模型的非凸性和噪声,非负矩阵分解方法容易陷入局部最优解的问题,提出一种再权重稀疏和正交约束非负矩阵分解算法(Reweight Sparse and Orthogonal Nonnegative Matrix Factorization, RONMF)。RSNMF是一种稀疏增强的算法,充分体现了高光谱图像解混的地物丰度稀疏性,但也因此使得光谱近似的地物容易混淆。RONMF在再权重稀疏非负矩阵分解的基础上,引入正交非负矩阵分解(Orthogonal Nonnegative Matrix Factorization, ONMF),增强端元光谱的独立性,在再权重稀疏算法基础上进一步优化,以达到更好的解混效果。实验也证实了该算法的优越性能,RONMF算法对土壤与路这种光谱相近的端元解混性能与SONMF相近,继承SONMF有效保护端元独立性的特性,对树和水这种丰度稀疏特性较强端元的解混性能,极大程度地保留了再权重稀疏算法的稀疏性增强能力。

The nonnegative matrix factorization model has the non convexity and noise, and it is easy to fall into local optimal solution. Aiming at these problem, we propose a reweight sparse and orthogonal nonnegative matrix factorization(RONMF). Reweight sparse nonnegative matrix factorization(RSNMF) is a enhancing sparse algorithm, which fully reflects the sparse abundance of ground objects of hyperspectral unmixing. However, the ground objects with similar spectrum is easily to be confused. On the basis of RSNMF, RONMF introduced orthogonal nonnegative matrix factorization(ONMF) to enhance the spectral independence of endmembers and further optimized the RSNMF to achieve better effect of unmixing. The experiment proved the superior performance of the proposed algorithm. RONMF has similar performance to SONMF on approximated endmembers such as solid and road, which inherits the characteristics of SONMF of effectively protecting the spectrum independence. It greatly retains the sparsity enhancement ability from RSNMF for endmembers which have a strong sparsity such as tree and water.