Using convolutional neural networks to predict galaxy metallicity from three-colour images

2018/10/31 by John F. Wu, Steven Boada · 45 citations
Mathematics · Physics and Astronomy · #Algorithm #Artificial intelligence #Astrophysics #Computer science #Convolutional neural network #Galaxies: Formation, Evolution, Phenomena #Galaxy #Gamma-ray bursts and supernovae #Mathematics #Mean squared error #Metallicity #Optics #Pattern recognition (psychology) #Physics #Pixel #Random forest #Residual #Sky #Statistics #Stellar, planetary, and galactic studies #astro-ph.GA

paper · pdf · doi:10.1093/mnras/stz333

published in Monthly Notices of the Royal Astronomical Society 484(4), 4683-4694 (Oxford University Press) · 13 pages, 6 figures, accepted to MNRAS. Code is available at https://github.com/jwuphysics/galaxy-cnns

openalex created_date 2018/11/09 · arxiv created 2019/01/30 · openalex publication_date 2019/01/31 · arxiv updated 2019/03/04 · openalex updated_date 2026/08/05

Abstract

We train a deep residual convolutional neural network (CNN) to predict the gas-phase metallicity (Z) of galaxies derived from spectroscopic information (⁠|Z ≡ 12 + log (\rm O/H)|⁠) using only three-band gri images from the Sloan Digital Sky Survey. When trained and tested on 128 × 128-pixel images, the root mean squared error (RMSE) of Zpred − Ztrue is only 0.085 dex, vastly outperforming a trained random forest algorithm on the same data set (RMSE = 0.130 dex). The amount of scatter in Zpred − Ztrue decreases with increasing image resolution in an intuitive manner. We are able to use CNN-predicted Zpred and independently measured stellar masses to recover a mass–metallicity relation with 0.10 dex scatter. Because our predicted MZR shows no more scatter than the empirical MZR, the difference between Zpred and Ztrue cannot be due to purely random error. This suggests that the CNN has learned a representation of the gas-phase metallicity, from the optical imaging, beyond what is accessible with oxygen spectral lines.

Citations

• Adam: A Method for Stochastic Optimization
• Evaluation of Pooling Operations in Convolutional Architectures for Object Recognition
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Improving neural networks by preventing co-adaptation of feature detectors
• Preparing Red-Green-Blue (RGB) Images from CCD Data
• A MACHINE LEARNING APPROACH FOR DYNAMICAL MASS MEASUREMENTS OF GALAXY CLUSTERS
• Matplotlib: A 2D Graphics Environment
• Rotation-invariant convolutional neural networks for galaxy morphology prediction
• On the histogram as a density estimator:L 2 theory
• The Origin and Evolution of the Mass-Metallicity Relationship for Galaxies: Results from Cosmological N -Body Simulations
• Backpropagation Applied to Handwritten Zip Code Recognition
• A Survey on Transfer Learning
• The Formation of Dust Lanes: Implications for Galaxy Evolution
• SGDR: Stochastic Gradient Descent with Warm Restarts
• The Fourteenth Data Release of the Sloan Digital Sky Survey: First Spectroscopic Data from the Extended Baryon Oscillation Spectroscopic Survey and from the Second Phase of the Apache Point Observatory Galactic Evolution Experiment
• Decoupled Weight Decay Regularization
• Deep Residual Learning for Image Recognition