[Previous] | [Session 32] | [Next]
S Erard, D Despan (Observatoire de Paris-LESIA)
Modern space borne instruments produce very large data sets that require automated processing and analysing methods. Imaging spectroscopy data may be particularly difficult to handle because they consist in very correlated data sets where the relevant information only bears a small fraction of the overall variance. For this reason, analyses using multivariate methods often require a preprocessing step to extract the useful signal (in this case bands center, width, and depth). Although such methods (such as MGM analysis, Sunshine et al. JGR 1990) exist and are helpful to some extend, they have drawbacks that limit they automatic application. In most cases, they require some a priori knowledge of the information contents e. g., the number of absorption bands and their approximate locations. For this reason, alternative methods are needed.
We present here a new method based on wavelet decomposition and on a multiscale vision model, partly derived from image analysis techniques (e.g., Starck et al. Ap. J. 1997). The outcome of the analysis is a description of the bands detected, and a quantitative and reliable confidence parameter. The band can be described either by the most appropriate wavelet scale only (for rapid analyses) or after reconstruction from all scales involved (for more precise measurements). The principle of the method is presented here, as well as analysis strategies. It is then tested on reference data sets including simulations, laboratory spectra of minerals (controlled olivine / pyroxenes mixtures), and planetary observations.
[Previous] | [Session 32] | [Next]
Bulletin of the American Astronomical Society, 37 #3
© 2004. The American Astronomical Soceity.