[A novel scheme SVR(HAAR) for automatically estimating stellar atmospheric parameters from spectrum]

Area of Science:

• Astronomy and Astrophysics
• Signal Processing
• Machine Learning

Context:

• Stellar spectra analysis is crucial for understanding celestial objects.
• Spectral data is often corrupted by noise from various sources, impacting parameter estimation.
• Existing methods struggle with noise and minor spectral distortions.

Purpose:

• To propose a novel SVR(Haar) scheme for robust and accurate estimation of physical parameters from stellar spectra.
• To leverage Haar wavelet transforms for noise reduction and feature extraction.
• To enhance estimation accuracy and robustness against spectral distortions.

Summary:

• A new SVR(Haar) method combines Haar wavelet feature extraction with support vector regression (SVR).
• Haar wavelets effectively remove high-frequency noise, improving data quality.
• The epsilon-insensitive domain in SVR enhances robustness to calibration imperfections.

Impact:

• The SVR(Haar) scheme demonstrates superior performance compared to traditional methods like Principal Component Analysis (PCA) and non-parametric regression.
• Validated on simulated and real stellar spectra (SLOAN data), proving its effectiveness.
• Offers a more accurate and reliable approach for astrophysical data analysis.