Abstract:Advances in the technology of astronomical spectra acquisition have resulted in an enormous amount of data available in world-wide telescope archives. It is no longer feasible to analyze them using classical approaches, so a new astronomical discipline, astroinformatics, has emerged. We describe the initial experiments in the investigation of spectral line profiles of emission line stars using machine learning with attempt to automatically identify Be and B[e] stars spectra in large archives and classify their types in an automatic manner. Due to the size of spectra collections, the dimension reduction techniques based on wavelet transformation are studied as well. The result clearly justifies that machine learning is able to distinguish different shapes of line profiles even after drastic dimension reduction.

Using data mining techniques applied on emission line characteristics of Be stars spectra we attempted to find new Be stars candidates in SDSS SEGUE survey. The mid-resolution spectra of confirmed Be stars obtained from VO-compatible archive of Ondřejov observatory 2m telescope were transformed to the spectral resolution of SDSS and important characteristics of emission line profiles were estimated, to be used as a training base of supervised learning methods. The obtained knowledge base of the characteristic shapes and sizes of Be emission lines was finally used to identify new potential candidates in SDSS spectral survey. The several newly found Be stars candidates justify our approach and approve Astroinformatics as a viable research methodology.