SUPPNet: Neural network for stellar spectrum normalisation

Abstract

Context. Precise continuum normalisation of merged échelle spectra is a demanding task that is necessary for various detailed spectroscopic analyses. Automatic methods have limited effectiveness due to the variety of features present in the spectra of stars. This complexity often leads to the necessity for manual normalisation which is highly time-consuming. Aims. The aim of this work is to develop a fully automated normalisation tool that works with order-merged spectra and offers flexible manual fine-tuning, if necessary. Methods. The core of the proposed method uses the novel, fully convolutional deep neural network (SUPP Network) that was trained to predict a pseudo-continuum. The post-processing step uses smoothing splines that give access to regressed knots, which are useful for optional manual corrections. The active learning technique was applied to deal with possible biases that may arise from training with synthetic spectra and to extend the applicability of the proposed method to features absent in this kind of spectra. Results. The developed normalisation method was tested with high-resolution spectra of stars with spectral types from O to G, and gives a root mean squared (RMS) error over the set of test stars equal to 0.0128 in the spectral range from 3900 Å to 7000 Å and 0.0081 in the range from 4200 Å to 7000 Å. Experiments with synthetic spectra give a RMS of the order of 0.0050. Conclusions. The proposed method leads to results that are comparable to careful manual normalisation. Additionally, this approach is general and can be used in other fields of astronomy where background modelling or trend removal is a part of data processing.

Melchiors

“…The procedure to pseudo-normalise the HERMES spectra was eased by the previous steps of the data reduction, but it remained challenging since the spectral range to normalise runs over more than 5000 Å. Manual pseudo-normalisation tools exist, but they render the normalisation process highly time-consuming and do not ensure the repeatability of the result. Recent development to make automatic and tunable normalisation processes has emerged based on the convex-hull and alpha-shape theories, such as AFS (Xu et al 2019) and RASSINE (Cretignier et al 2020), and even based on machine learning techniques, for example, using deep neural networks (Różański et al 2022). In this work, we rely on a more classical approach based on a iterative filtering method similar to the ones that use a sigmaclipping process, as done, for example, on a smaller wavelength extent in Muñoz Bermejo et al ( 2013 We provide normalised spectra when the best d score is lower than 0.24.…”

Artificial Intelligence in Astronomical Optical Telescopes: Present Status and Future Perspectives

“…KNN [151], PCA, and Bayesian methods [152][153][154] have been used to effectively estimate stellar physical parameters. With the sudden progress of AI, different machine learning methods have been utilized to analyze stellar parameters [155][156][157][158], and some powerful pipeline tools based on machine learning have been implemented for stellar physics parameter estimation and measurement, such as ODUSSEAS [159], ROOSTER [160], and SUPPNet [161]. Benefiting from a huge volume of astrophysical data, deep learning methods are also widely used in this field and have become a research hotspot [162][163][164], and the assessment of stellar physical parameters yields remarkable results.…”

Prospects for future studies using deep imaging: Analysis of individual Galactic cirrus filaments

“…This type of network architecture is successfully applied to various scientific and applied tasks such as medical image analysis (Iglovikov et al 2017b;Ching et al 2017;Ing et al 2018a;Ing et al 2018b;Andersson et al 2019;Nazem et al 2021), cell biology (Kandel et al 2020), and satellite image analysis (Iglovikov et al 2017a). It is also used in astronomical applications such as denoising, enhancing astronomical images (Vojtekova et al 2021), and stellar spectrum normalization (Różański et al 2022). In this section, we consistently describe these neural network models, through datasets (Sect.…”