Rare Object Search from Low-S/N Stellar Spectra in SDSS

• Published in: IEEE access Vol. 8; p. 1
• Main Authors: Wu, Minglei, Pan, Jingchang, Yi, Zhenping, Wei, Peng
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01-01-2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Astronomy; Carbon; Carbon stars; Cataclysmic variables; Clustering; Density; Feature extraction; Galactic evolution; Galaxies; Machine Learning; Object recognition; Outliers (statistics); Principal component analysis; Principal components analysis; Random forests; Random noise; Rare Object Search; SDSS; Search problems; Searching; Sky surveys (astronomy); Spectral classification; Stellar Spectra; Support vector machines; Wavelength division multiplexing; White dwarf stars
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.2983745

Rare objects such as white dwarf+main sequence (WDMS) and cataclysmic variables (CVs) are very important for studying the evolution of the galaxy and the universe. The large amount of spectra obtained by the large sky surveys such as the Sloan Digital Sky Survey (SDSS) are rich sources of these rare objects. However, a considerable fraction of these spectra are low-S/N spectra. These low-S/N spectra contain similar useful information as the high-S/N spectra, and making better use of these spectra can significantly improve the chance of finding rare objects. Nevertheless, little research has been done on them. In this study we propose a novel method based on the combination of PCA (Principal Components Analysis) and CFSFDP (Clustering by Fast Search and Find of Density Peak) to search for rare objects from low-S/N spectra. The PCA first extracts principal components from high-S/N spectra to generate general feature spectra and reconstructs low-S/N stellar spectra with these general feature spectra. Then the CFSFDP calculates the Local Density ρ and the Distance δ of the reconstructed spectra, and select the outliers through the decision graph quickly and accurately. We first apply our method to spectra in SDSS stellar classification template library with adding white gaussian noise to search for rare objects (carbon stars, carbon white dwarfs, carbon_lines, white dwarfs and white dwarfs magnetic). Then we apply our method to observed spectra with different low-S/Ns from SDSS and compared with Lick-index+K-means and Support Vector Machines (SVM). The experimental results show that our method has a higher efficiency compared to other methods.