Abstract

The Large Sky Area Multi-Object Fibre Spectroscopic Telescope (LAMOST) is producing a very large database which consists of spectra from the Chinese 4-meter telescope. LAMOST Data Release 3 (DR3 http://dr3.lamost.org/) contains over 5X10⁷ spectra, of which 5.4X10⁶ are stellar spectra. In this preliminary work we have used only the F, G and K spectral types of stars with S/N > 30. Our data set consists of 286,283 spectra that have been classified using Automated Supervised Neural Network (ANNs); this forms the test set, while the training set consists of spectra from the MILES spectral library. Of the three ANN tools used (Tree, Forest and Neural), the best performance was seen for the Tree based classifier; it returned a classification accuracy of 71.4% correct spectral types and an error of 4.61% spectral sub-types. The luminosity classes are not known for the vast majority of LAMOST spectra, but our automated schemes provide this information along with a confidence estimate (with corresponding classification probabilities) for each spectrum. Future work will involve (i) using more classification tools, (ii) improving the classification accuracies, and (iii) applying these upgrades to future LAMOST data releases.

Fig. 1	Fig. 2
Fig. 3	Fig. 4