Capturing star formation activity from compressed photometric images of galaxies

TL;DR

This paper introduces a machine learning method using Vision Transformers to classify star-forming galaxies directly from photometric images, eliminating the need for spectroscopic analysis, and is scalable for large astronomical surveys.

Contribution

The novel approach applies Vision Transformer models to classify galaxy star formation activity solely from JPEG images, bypassing traditional spectroscopic diagnostics.

Findings

Achieved accurate classification of star-forming galaxies from images.

Demonstrated method's scalability for large survey data.

Potential for application in upcoming astronomical surveys.

Abstract

We present a novel approach for classifying star-forming galaxies using photometric images. By utilizing approximately $124,000$ optical color composite images and spectroscopic data of nearby galaxies at $0.01<z<0.06$ from the Sloan Digital Sky Survey, along with follow-up spectroscopic line measurements from the OSSY catalog, and leveraging the Vision Transformer machine-learning technique, we demonstrate that galaxy images in JPEG format alone can be directly used to determine whether star-forming activity dominates the galaxy, bypassing traditional spectroscopic analyses such as emission-line diagnostic diagrams. We anticipate that this method holds significant potential for application in current and future large-scale surveys, such as Euclid, the Dark Energy Survey (DES), and the Legacy Survey of Space and Time (LSST).