期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2019
卷号：46
期号：2
页码：291-299
出版社：IAENG - International Association of Engineers
摘要：Hyperspectral unmixing (HU) has been widely used to address the mixed-pixel problem in the quantitative analysis of hyperspectral remote sensing images, in which endmember extraction plays a very important role. In this paper, a two-stage algorithm is presented for endmember extraction. The first stage aims at finding pure endmembers that have pure pixel representations in the hyperspectral scene. At first, pure-pixel-based endmember extraction algorithms are exploited to find the spectrally pure pixels directly from hyperspectral images as initial pure pixel candidates, and then local spatial-spectral information is utilized to determine pure endmembers. The second stage aims at generating virtual endmembers (not necessarily present in the set comprised by input data samples). We extend the original nonnegative matrix factorization (NMF) unmixing model to incorporate endmember a priori information, and then use the extended NMF method to generate virtual endmembers. Experimental results with both simulated and real hyperspectral data sets have validated the effectiveness of our method and have demonstrated that the known endmember information is beneficial to the extraction of other unknown endmembers.
