Automatic Normalization Method of Stellar Spectrum Based on Spline Function

The observed spectrum of stars is generally composed of continuous spectrum, spectral line and noise. The continuous spectrum is the smooth continuous spectrum of radiation flux varying with wavelength caused by blackbody radiation. Spectral classification and stellar physical parameter estimation depend on the accurate extraction of continuous spectrum and spectral line information. Therefore, the main work of spectral data processing is to fit continuous spectrum and extract spectral line features by normalization. At present, the methods of continuous spectrum fitting mainly include polynomial fitting, median filtering and wavelet filtering. Existing methods have limitations to varying degrees in the case of low SNR, interference of cosmic ray signals and existence of emission lines, which are mainly reflected in robustness and accuracy. For the time being, there is no automated method applying to the normalization of the 10⁷ spectra from LAMOST. In the period of avalanche of astronomical data, it is very urgent to research and develop a spectral normalization algorithm of stars with better universality and automatic processing that can be applied to a wider range of temperature, SNR and wavelength coverage. On the basis of careful analysis of different types of spectra, a continuous spectrum fitting method based on fixed window dividing is proposed. This method can filter and extract the data points in the spectrum which can reflect the characteristics of continuous spectrum,and produce more accurate continuous spectrum by fine controlling the smoothness of spline function. Experiments are carried out using spectra of different spectral types, temperature ranges and wavelength coverage ranges in LAMOST, and the results show that the proposed algorithm has good accuracy and universality.