DAHLIA

Hyperspectral imaging is popular in many science areas such as remote sensing or laboratory experiments. In astronomy its spectacular development in the context of present (8m class with adaptive optics) and future (e.g. ELT and JWST) telescopes will provide soon to the community hyperspectral data with an unprecedented level of complexity. With its 90,000 spatial elements and 4,000 spectral channels, the 2nd generation AO assisted integral field spectrograph MUSE for the VLT is representative of these new instruments which will provide massive hyperspectral data. Present existing data analysis tools cannot cope with such data volume and characteristics. In order to achieve the best science return with these facilities we propose to develop innovative signal processing methods in the following areas: (1) advanced visualization of hyperspectral data, (2) optimal fusion of hyperspectral data for increasing the field of view (mosaicing) or improving the S/N (summation) including a proper handling of the noise properties (using probabilistic and inverse problem approaches), (3) hyperspectral deconvolution taking into account a partially known PSF which is function of field location, wavelength and time (none of the currently available methods are efficient in such a context), (4) automatic source extraction in dense fields (studying Sparce Component Analysis and Non-Negative Matrix Factorization methods), (5) blind source localization and characterization in deep fields (cf. data mining, unsupervised segmentation, metrics definition, information theory, classification). To achieve these ambitious goals we have gathered a multi-disciplinary experienced group of IT scientists and astrophysicists from 4 institutes. The astrophysicists will help to precise the requirements of each task and will provide simulated and real data to validate the methods. The IT scientists will select and develop the most appropriate method to meet the requirements. The group has not only a world leading expertise in instrumentation (CRAL, LATT) and in signal processing for astronomical application (LATT, SIIT, OCA) but is also experienced in managing large projects. This development will be carried in the context of the MUSE project in which CRAL and LATT are respectively leader and member of the European consortium of 7 institutes which build the instrument. The produced algorithms will then result in a strong asset for the French leadership on the project and will enhance the scientific value of the 255 VLT nights of guarantee time which have been granted to the consortium. While the first three years (2008-2010/2011) are devoted to the development and prototyping of the new algorithms using sophisticated simulations, the last year (2011-2012) will be focused on production software development and final validation on real MUSE data. The resulting methods will be published in both IT and astrophysics journals and the software will be made public through the Virtual Observatory. An international workshop will also be organized in view of disseminating the results produced in this study.