An artificial neural network approach to automatic classification of stellar spectra

Abstract

This paper presents the design and implementation of several models of artificial neural networks for the automatic classification of low-resolution spectra of stars. In previous works, we have developed knowledge-based systems for the analysis of spectra. We shall now use these analysis methods to extract the most important spectral features, training the proposed neural networks with this numeric characterization. Although there are published works about neural networks applied to the classification problem, our final purpose is the integration of several artificial techniques in a unique hybrid system. In the development of such a system we have combined signal processing techniques, knowledge-based systems, fuzzy logic and artificial neural networks, integrating them by means of a relational database which allow us to structure the collected astronomical data and also contrast the results achieved with the different classification methods. © Springer-Verlag Berlin Heidelberg 2003.