ACCURATE 2-DIMENSIONAL CLASSIFICATION OF STELLAR SPECTRA WITH ARTIFICIAL NEURAL NETWORKS

We present a solution to the long-standing problem of automatically cl assifying stellar spectra of all temperature and luminosity classes wi th the accuracy shown by expert human classifiers. We use the 15 Angst rom resolution near-infrared spectral classification system described by Torres-Dodgen & Weaver in 1993. Using the spectrum with no manual i ntervention except wavelength registration, artificial neural networks (ANNs) can classify these spectra with Morgan-Keenan types with an ac curacy comparable to that obtained by human experts using 2 Angstrom r esolution blue spectra, which is about 0.5 types (subclasses) in tempe rature and about 0.25 classes in luminosity. Accurate temperature clas sification requires a hierarchy of ANNs, while luminosity classificati on is most successful with a single ANN. We propose an architecture fo r a fully automatic classification system.