基于OSP与NMF的光谱混合像元分解方法

The Spectral Unmixing Based on OSP and NMF

非负矩阵分解(NMF)由于跟线性光谱混合模型具有很高的相似性,因此成为光谱混合像元分解中算法中的一个研究热点。为了避免NMF算法陷入局部最小带来的求解结果不确定性,提出用正交子空间投影(OSP)方法来估计高光谱图像端元的个数,同时简化了最小单形体体积约束的NMF算法中关于单形体体积的计算方法。实验结果表明利用该算法得到的地物丰度图与真实地物的分布状况相吻合。

Because of the high similarity to the linear spectral mixing model,non-negative matrix factorization（NMF）has become a hot research topic in spectral unmixing algorithms.In order to avoid uncertainty of the results due to the NMF algorithm dropping into a local minimum solution,this paper proposes orthogonal subspace projection（OSP）to estimate the number of endmembers,Meanwhile,the calculation method for the simplex volume is simplified.The experimental results show that the abundance map obtained from spectral unmixing can reflect the real distribution of surface minerals.