最小体积和平滑性约束的非负矩阵分解高光谱解混算法

Hyperspectral unmixing base on minimum volume and smoothes constrained nonnegative matrix factorization

针对传统非负矩阵分解算法约束项中仅考虑了端元或丰度的相关属性,存在着解混精度欠佳、对噪声鲁棒性差等问题,提出了一种基于最小体积和平滑性约束的非负矩阵分解高光谱图像解混算法。首先利用了高光谱图像端元的几何特性,同时考虑了丰度的平滑特性,并将这两种特性结合成约束项,同时加入到了非负矩阵分解的目标函数中。然后,通知优化目标函数的求解方法,计算出端元和丰度。实验结果表明,不管是合成的模拟图像还是真实的高光谱场景图像,新提出的方法解混结果都能够满足精度要求。

Traditional Nonnegative Matrix Factorization(NMF)methods only take endmember or anbundance into consideration,leading to unmixing results and poor robustness with noise while applying for hyperspectral unmixing.A novel minimum volume and smoothess constrained NMF were proposed in the paper.Firstly,the geometry properties of endmember and smoothness of abundance were combined in the objective function of NMF.Then,getting the endmember and aubndance by optimizing the calculate method of objective function.The experiment on synthetic and real hyperspectral images showed that the method proposed could meet the accuracy requirements.