一种有效的无监督高光谱图像解混方法

摘要

The traditional nonnegative matrix factorization (NMF) involves the extraction of endmenber and the estimation of abundance values for each endmember when beding applied to hyperspectral unmixing.The random initialization for unmixing affect the results.In ordor to solve the problem,a new approach based on NMF which contain initialization was proposed for decomposition of mixed pixels by improving the way of getting initialization.Without a priori studies,the actual unmixing is performed by repeatedly calculate the source dataset to get the initialization while estimate the endmembers.Experimental results on simulated and real hyperspectral data demonstrate that the proposed approach is effective.%传统的非负矩阵分解方法应用在高光谱像元分解时,混合像元分解用到的初始值是随机产生的,会影响像元的解混效果.为了解决该问题,本文通过改进获得初始值的方法,提出了一种基于改进非负矩阵分解的高光谱图像解混方法.本方法不需要任何先验知识,只需要对输入数据集进行迭代计算得到合适的初始值,即可达到光谱解混的目的.仿真数据和真实数据的实验结果都证明了该算法的有效性.