An Automated Stellar Spectra Classification System Based on Non-Parameter Regression and Nearest Neighbor Method

The automated classification and recognition of stellar spectra is an important research for the spectra processing system of modern telescope survey project. For the spectra without flux calibration. the authors present all automated stellar spectra classification system to achieve two goals: one is the spectral class and spectral subclass classification, and the other is luminosity type recognition. The system is composed of three units: (1) continuum normalization method based on wavelet technique; (2) non-parameter regression method for spectral class and spectral subclass classification; (3) chi(2) method based oil nearest neighbor for luminosity type determination. The experiments on low-resolution spectra show that the system achieves 3.2 spectral sub-class precision for spectral and spectral subclass classification, 60% correct rate for luminosity recognition, and 78% rate for the luminosity recognition with error less than or equal to 1. The system is easy, rapid in training, and feasible for the automated spectra classification.