Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation

Details Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation

Dec 2010

adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2010agufm.p53b1518t&link_type=abstract

American Geophysical Union, Fall Meeting 2010, abstract #P53B-1518

Statistics

Computation

[5464] Planetary Sciences: Solid Surface Planets / Remote Sensing, [6225] Planetary Sciences: Solar System Objects / Mars, [6295] Planetary Sciences: Solar System Objects / Venus

Scientific paper

Spectro-imagers with improved spectral/spatial resolution have mapped planetary bodies, providing high dimensional hyperspectral datasets that contain abundant data about the surface and/or atmosphere. The spatial extent of a pixel is usually large enough to contain a mixture of various surface/atmospheric constituents which contribute to a single pixel spectrum. Unsupervised spectral unmixing [1] aims at identifying the spectral signatures of materials present in the image and at estimating their abundances in each pixel. Bayesian Positive Source Separation (BPSS) [2] is an interesting way to deal with this unmixing challenge under linearity constraints. Notably, it ensures the non-negativity of both the unmixed component spectra and their abundances. Such a constraint is crucial to the physical interpretability of the results. A sum-to-one constraint [3] can also be imposed on the estimated abundances: its relevance depends on the nature of the dataset under consideration. Despite undeniable advantages, the use of such algorithms has so far been hampered by excessive computational resource requirements; so far it has not been possible to process a whole hyperspectral image of a size typically encountered in Earth and Planetary Sciences. Two kinds of implementation strategies were adopted to overcome this computational issue [4]. Firstly, several technical optimizations made it possible to run the BPSS algorithms on a complete image for the first time. Secondly, a pixel selection method was investigated: performed as a preprocessing step, it aims at extracting a few especially relevant pixels among all the image pixels. Then, the algorithm can be launched on this selection, with significantly lower computation overhead. In order to better understand the behavior of the method, tests on synthetic datasets generated by linear mixing of known mineral endmembers were performed. They help to assess the potential loss of quality induced by the pixel selection, depending on various characteristic parameters, e.g. number of endmembers, noising, pixel purity limitations. The method was then applied on real planetary datasets. We firstly analyzed data from the OMEGA instrument [5] (Mars Express) which yields information about Martian surface mineralogy. Convincing results were obtained for an image of the South polar cap of Mars [4]. We also analyzed data from the VIRTIS instrument [6] (Venus Express) which provides spectral signatures relative to the dense Venusian atmosphere. Sources corresponding to distinct atmospheric layers seem to be identified. Overall, results gathered in those different cases prove the ability of the BPSS approaches to provide a pertinent insight in the analysis of different planetary hyperspectral datasets. Ref.: [1] Keshava (2003), Lincoln Lab. Journal, no.14. [2] Moussaoui et al. (2008), Neurocomputing, vol.71, no.10. [3] Dobigeon et al. (2009), Signal Processing, vol.89, no.12. [4] Schmidt et al. (2010), accepted in IEEE/TGRS, http://arxiv.org/abs/1001.0499. [5] Bibring et al. (2004), ESA SP-1240. [6] Piccioni et al. (2006), ESA SP-1291.

Affiliated with

Also associated with

No associations

Rating

Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Investigating Martian and Venusian hyperspectral datasets through Positive Source Separation will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFWR-SCP-O-1499235