Endmember Extraction Algorithm Based on RMS Error Analysis in Hyperspectral Imagery

李智勇,匡纲要,薛绮

DOI: https://doi.org/10.3969/j.issn.1004-0323.2005.02.013

2005-01-01

Abstract:The linear mixed mixture is a commonly accepted model for hyperspectral data processing. As the important parameter of the linear mixed mixture, endmember represents a certain ground component whose spectral is changeless comparatively. Over the past years, several techniques have been developed for accomplishing the determination of the endmembers in the literatures. Aimed at the disadvantages of these existing methods, this paper presents an automatic endmember extraction algorithm based on RMS error analysis. The new algorithm overcomes the insufficiency of other methods and needs less prior information. Not only the mean spectrum of the imagery is extracted, but also the anormaly is reserved. This algorithm also depends on the linear mixed mixture, it is executed directly on the data with no previous dimensionality reduction. The mean spectrum of the data is chosen to start the process, and then, a series of constrained unmixing operations is performed, each time selecting as endmembers the pixels that minimize the remaining RMS error in the unmixed image. This process is continued until the required number of endmembers is found or the unmixing error is below some threshold. The results of simulated and real hyperspectral data show the validity of the algorithm by a detailed comparative analysis with PPI.