[A new method of sparse feature extraction for stellar spectra].

Area of Science:

• Astronomy and Astrophysics
• Data Science
• Signal Processing

Context:

• Stellar spectra parameterization is crucial for understanding stellar evolution and properties.
• Existing methods for feature extraction can be computationally intensive and prone to noise.
• Accurate estimation of atmospheric parameters (effective temperature, surface gravity, metallicity) is fundamental in astrophysics.

Purpose:

• To develop a novel, efficient, and accurate feature extraction method for stellar spectra.
• To improve the estimation of stellar atmospheric parameters using a combination of Haar wavelet and Lasso algorithm.
• To validate the proposed method on a large dataset of stellar spectra.

Summary:

• Stellar spectra are processed using multi-scale Haar wavelet decomposition to remove high-frequency noise.
• The Lasso algorithm is employed to select optimal features, reducing data dimensionality and redundancy.
• A non-parametric regression model utilizes these optimal features to estimate atmospheric parameters: log Teff, log g, and [Fe/H].

Impact:

• The Haar+lasso method significantly improves the accuracy and efficiency of stellar atmospheric parameter estimation.
• Achieved mean absolute errors of 0.0071 dex for log Teff, 0.2252 dex for log g, and 0.1996 dex for [Fe/H] on SDSS spectra.
• This novel approach offers a more precise alternative for deriving stellar atmospheric parameters compared to existing literature.