Stellar classification has long been a useful tool for probing important astrophysical phenomena. Beyond simply categorizing stars it yields fundamental stellar parameters, acts as a probe of galactic abundance distributions and gives a first foothold on the cosmological distance ladder. The MK system in particular has survived on account of its robustness to changes in the calibrations of the physical parameters. Nonetheless, if stellar classification is to continue as a useful tool in stellar surveys, then it must adapt to keep pace with the large amounts of data which will be acquired as magnitude limits are pushed ever deeper. We are working on a project to automate the multi-parameter classification of visual stellar spectra, using artificial neural networks and other techniques. Our techniques have been developed with 10,000 spectra (B <= 11) extracted from the objective prism plates of the Michigan Blue Survey, as well as follow-up CCD objective prism spectra of fainter stars (B <= 14). In addition to classification using the whole spectrum over the MK range, we have investigated the use of Principal Components Analysis as a front-end compression of the data. Our continuing work also looks at the application of synthetic spectra to the direct classification of spectra in terms of the physical parameters of Teff, log g, and [Fe/H].
American Astronomical Society Meeting Abstracts #188
- Pub Date:
- May 1996