Previous
abstract 
Next
abstract
Session 19 - Hot Stars.
Display session, Monday, January 15
North Banquet Hall, Convention Center
We have designed a suite of Artificial Neural Networks (ANNs) that are able to classify accurately the near-infrared spectra of ordinary stars in temperature and luminosity. The temperature classes O - M and luminosity classes Ia - V are classified in the extension of the MK System in the silicon-accessible (peak CCD and Reticon sensitivities) near-infrared (\lambda\lambda 5800 -- 8900). This classification system (Torres-Dodgen and Weaver, 1993, PASP, 105, 693) is rigorously based on MK standards.
Although we employ only 15 Åresolution spectra, the ANNs achieve fractional sub-class temperature and luminosity accuracies comparable to those obtained by expert classifiers with 2 Åresolution photographic spectra. Even OB stars are well classified by this system. We have demonstrated previously (Weaver and Torres-Dodgen, 1995, Ap.J. 446, 300 and references therein) that ANN classification can detect, and classify the components of, binary stars; can accurately determine interstellar reddening during classification; and degrades slowly with decreasing signal-to-noise ratios. The remaining issue is to determine the limit of spectral peculiarities that can be detected by ANNs at these resolutions.
Such ANNs can be used for automated spectroscopic surveys that are as accurate but two magnitudes deeper than spectroscopic surveys using expert human classifiers.
