Deriving Stellar Parameter Using Radial Basis Function Neural Network

The large sky area multi-object fiber spectroscopic telescope (LAMOST) will collect enormous amounts of low-resolution spectra on a significant fraction of the stellar content of our Galaxy. Various reliable and automated techniques are required to determine physical parameters from such a huge set of stellar spectra. In this paper, we presents a method that employs radial basis function (RBF) neural network to derive the stellar parameters namely effective temperature T_eff, metallicity [Fe/H] and gravity logg from stellar spectra. Studying those theoretical spectra of which the features are defined by several line indices, we trained a RBF neural network through a part of a theoretical spectra collection to set up the relationship between the line indices and the stellar parameters. The network was then applied to predict physical parameters for another part of the data set. The experiment result shows that within the range of temperature from 4000 to 10000 K and the metallicity from -4.0 to 0.5 dex (in logarithm), statistical errors of T_eff and [Fe/H] are smaller than 168 K and 0.28 dex respectively, and the error of logg is smaller than 0.3 dex within the range from 0 to 5.0 dex. Such precision is practicable for the statistical research of our Galaxy.