SUPPNet: Neural network for stellar spectrum normalisation

TL;DR

This paper introduces SUPPNet, a deep neural network-based tool for automated stellar spectrum normalisation that achieves accuracy comparable to manual methods and can be fine-tuned manually if needed.

Contribution

The paper presents a novel fully convolutional neural network for spectrum normalisation, incorporating active learning and post-processing for improved accuracy and flexibility.

Findings

Achieves RMS error of 0.0128 in real spectra

Comparable to manual normalisation results

Effective on synthetic spectra with RMS of 0.0050

Abstract

Precise continuum normalisation of merged \'{e}chelle spectra is a demanding task 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 of manual normalisation which is a time demanding task. 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. 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 gives access to regressed knots 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. The developed normalisation method was tested with high-resolution spectra of stars having spectral types from O to G, and gave root mean squared (RMS) error over the set of test stars equal $0.0128$ in the spectral range from $3900\,\r{A}$ to $7000\,\r{A}$ and $0.0081$ in the range from $4200\,\r{A}$ to $7000\,\r{A}$. Experiments with synthetic spectra give RMS of the order of $0.0050$. The proposed method gives results 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. The algorithm is available online at https://git.io/JqJhf.