Preparing for LSST: testing deep learning pipelines for galaxy merger classification

Dey, Subhrata; Pearson, William; Cotter, Aidan

doi:10.5281/zenodo.17328233

Preparing for LSST: testing deep learning pipelines for galaxy merger classification

Dey, Subhrata; Pearson, William; Cotter, Aidan

Publisher: Zenodo
DOI: 10.5281/zenodo.17328233
Repository: zenodo.org
Language: en

Abstract

Galaxy mergers are fundamental to the hierarchical assembly and evolution of galaxies, often driving starburst activity and AGN fueling. Identifying mergers and their stages, such as pre- and post-coalescence, from imaging alone, especially given the vast size of modern datasets, remains extremely challenging. We develop a supervised deep learning framework using Convolutional Neural Networks (CNNs) to classify galaxies as non-mergers, pre-mergers, or post-mergers. Our training uses mock Hyper Suprime-Cam (HSC) images from the IllustrisTNG simulations (Margalef-Bentabol et al. 2024). HSC, a precursor to LSST, is ideal for developing and validating machine learning methods for future surveys. We test our model on synthetic and real HSC data to assess robustness and generalizability. This approach demonstrates the potential of simulation-driven machine learning to reveal galaxy merger histories in upcoming wide-field surveys.

Full text

Preparing for LSST: testing deep learning pipelines for galaxy merger classification Subhrata Dey William Pearson, Aidan Cotter National Centre for Nuclear Research, Warsaw, Poland Taffy Galaxies colliding NOIRLAB The mice NGC 6240 Why galaxy mergers are important? ●Shapes the galaxies – mergers alter galaxies’ structure, size, and dynamics. ●Triggers star formation – interactions compress gas clouds, triggering bursts of star formation. ●Fuel AGN activity – mergers channel gas to galactic centers, fueling active nuclei. ●Redistributing material – stars, gas, and metals are mixed and spread across galaxies. Why galaxy mergers are important? ●Galaxy Growth – hierarchical merging is a key pathway for building massive galaxies. ●Tracing Cosmic Evolution – merger rates reﬂect how structure forms across cosmic time. Mergers are fundamental drivers of galaxy evolution. Identifying Merger Visual Classiﬁcation ESO Identifying Merger Humans are bound to make errors Close Pairs Galaxies close on the sky (< 20–50 kpc) Small relative line-of-sight velocity diﬀerence (e.g. ≤ 500–1000 km/s) Identifying Merger Morphological statistics Identifying Merger Morphological statistics Identifying Merger Cotter et al. in prep Workflow for Galaxy Merger Classification z= 0.10–0.31 0.31–0.52 0.52–0.76 0.76–1.00 Workflow for Galaxy Merger Classification Original Rotated Flipped Cropped (100 Kpc) Scaled Workflow for Galaxy Merger Classification Original Rotated Flipped Cropped (100 Kpc) Scaled ●Zoobot ConvNeXt ●EfficientNet-B3 ●ResNet-18 ●Swin Transformer (SwinT) ●Vision Transformer (ViT) Ensemble Method ●Combined predictions from all models ●Final class chosen only if confidence > 0.7. Classifier performance Accuracy: Overall correctness of the model Precision: Measures how reliable positive predictions are. Recall: Measures how completely the model identiﬁes positives. Classifier performance Ensemble performance vs redshift Ensemble performance vs mass Fraction of correctly classified Mergers vs Merger times Pre-mergers Ongoing-mergers post-mergers ●Trained 5 deep learning networks for merger stage classification -> Ensemble approach gave the best performance ●Reliable for non-mergers and pre-mergers ●Challenge: ongoing and post-mergers ○Morphologies evolve to resemble non-mergers ○Faint tidal features hinder classification Summary and Future outlook Outlook ●Build a comprehensive Non / Pre / Ongoing / Post-merger catalog for the NEP field ●LSST’s depth & sensitivity will uncover faint signatures → robust classification of all merger stages Thank You :) [email protected] Work in Progress!!!