Abstract—Spectral unmixing consists in finding a set of spectrally pure components (endmembers) and their corresponding fractions coverage for each pixel (abundances) in the hyperspectral data. Most the existing approaches consider the number of endmembers as input to their algorithms. In this paper, we propose a new mixture analysis method which relies on a spectral summarization algorithm that is inspired from convex geometry modelling, and works directly on the whole bandwidth range. It learns the endmembers and the abundances based on the information provided by the spectral summarization of the original scene. Also, it automatically optimizes the number of endmembers required by a particular scene. This optimization is achieved using the Competitive Agglomeration clustering algorithm.
Index Terms—hyper-spectral data, mixture analysis, unmixing
Cite: Ouiem Bchir, "Self-Estimation of the Number of Endmembers for Hyperspectral Mixture Analysis," Journal of Image and Graphics, Vol. 3, No. 1, pp. 56-59, June 2015. doi: 10.18178/joig.3.1.56-59
Copyright © 2012-2024 Journal of Image and Graphics, All Rights Reserved