Spectral Classification and Particular Spectra Identification Based on Data Mining

Spectral classification and particular spectra identification are primary tasks for celestial object study in astronomy. With the developing of large spectrographic surveys, huge volumes of spectral data can be easily obtained and various data mining methods have been widely applied to assist astronomers for automatic spectral analysis. In this paper, we review these methods in detail and analyze their advantages as well as disadvantages. Moreover, experimental results of the representative methods are reported and discussed from different perspectives, including number of training and testing samples, feature extraction scheme, classifier selection and performance evaluation. Finally, we point out the existing problems and the potential research trend.