Robust unmixing algorithms for hyperspectral imagery

Abderrahim Halimi, Yoann Altmann, Gerald Stuart Buller, Stephen McLaughlin, William Oxford, Damien Clarke, Jonathan Piper

Research output: Contribution to conferencePaperpeer-review

105 Downloads (Pure)


The linear mixture model (LMM) assumes a hyperspectral pixel
spectrum to be a linear combination of endmember spectra corrupted
by additive noise. This model is widely used for spectral
unmixing mainly because of its simplicity. However, the LMM
can be inappropriate in presence of nonlinear effects, endmember
variability or outliers. This paper presents a comparison between recent
robust hyperspectral unmixing algorithms. The mixture models
are first introduced followed by the description of their associated
unmixing algorithms. The algorithms are then analyzed when
considering a real image acquired over the region of Porton Down
in England. The results discuss the behavior of each algorithm to
unmix these data and compare their ability to detect the natural or
man-made outliers in the scene. The obtained results highlight the
potential of the studied mixture models to overcome the current
limitations of the LMM.
Original languageEnglish
Publication statusAccepted/In press - 16 Jun 2016
Event6th Sensor Signal Processing for Defence Conference 2016 - Edinburgh, United Kingdom
Duration: 22 Sept 201623 Sept 2016
Conference number: 6th


Conference6th Sensor Signal Processing for Defence Conference 2016
Abbreviated titleSSPD 2016
Country/TerritoryUnited Kingdom


Dive into the research topics of 'Robust unmixing algorithms for hyperspectral imagery'. Together they form a unique fingerprint.

Cite this