Main types of galaxy: How AI is Transforming Their Study and Classification

Title
How Artificial Intelligence is Transforming the Study and Classification of Galaxy Types

DOI: https://doi.org/10.5281/zenodo.17492307
Authors:
Nohil Kodiyatar* ORCID: https://orcid.org/0000-0001-8430-1641
Abhay Shamala ORCID: https://orcid.org/0009-0005-3261-8811
*Corresponding Author: Nohil Kodiyatar

---

Abstract

Galaxy morphology has long served as a key tool for understanding galaxy formation and evolution. The principal types of galaxies—elliptical, spiral (normal and barred), lenticular, irregular, peculiar (including interacting/merger systems), dwarf, and active nuclei hosts—offer insight into star-formation histories, dynamical processes, and environmental influences (Buta, 2010; Conselice, 2006). As modern sky surveys generate ever-larger image and spectroscopic data sets, traditional manual classification approaches become infeasible. In response, artificial intelligence (AI), especially machine learning (ML) and deep learning (DL) techniques, have been increasingly applied to automate classification, detect subtle structural features, and model evolutionary pathways. This paper presents a review of how AI is transforming galaxy classification: we summarise the main galaxy types, outline the AI methods applied, present recent results, and discuss challenges and future directions. We show that AI tools are enabling higher accuracy, greater depth (e.g., sub-types, multi-wavelength data), and expanded scale (millions of galaxies) in morphological analyses. We conclude by arguing that the integration of AI with traditional astrophysical interpretation is essential to leverage the next generation of surveys (e.g., LSST, Euclid) and to deepen our understanding of galaxy evolution.

Keywords: galaxy morphology; machine learning; deep learning; galaxy classification; artificial intelligence; galaxy evolution

---

Introduction

The morphology of galaxies—defined as their structural form, stellar content, gas/dust distribution, and dynamical state—provides a foundational probe into their origin and evolution (Buta, 2010; van den Berg, 1998). Classification schemes, such as the Hubble sequence, divide galaxies into major types (elliptical, lenticular, spiral, irregular) and thereby correlate morphology with star-formation history, environment, and dynamical processes (Conselice, 2006; Buta, 2010). Yet, with the increasing volume and complexity of astronomical data from modern surveys, traditional visual classification has become impractical for comprehensive studies (Lintott et al., 2011). In parallel, the rise of AI—particularly ML and DL methods—offers new pathways to automate classification, extract deeper morphological information, and uncover subtle patterns across vast data sets (Dieleman et al., 2015; Zhang et al., 2019).

This paper addresses how AI is transforming the study and classification of galaxy types. In the first section we summarise the principal galaxy types and their astrophysical importance. In the second, we review AI methods applied in morphology studies. In the third section, we examine how these methods have been applied to specific galaxy types (elliptical, spiral, lenticular, irregular, peculiar/merger, dwarf, active). Finally, we discuss the key challenges, opportunities, and future directions for integrating AI and galaxy-morphology studies.

---

Main Discussion / Analysis

Galaxy Morphology: Major Types and Physical Significance

Elliptical galaxies are characterised by smooth, spheroidal light distributions, dominated by old stellar populations and minimal ongoing star formation, with little gas and dust content (van den Berg, 1998). They often reside in dense environments such as galaxy clusters, and their structural properties implicate merger processes and dynamical relaxation (Conselice, 2006).
Spiral galaxies, including normal and barred sub-types, feature a central bulge, rotating disc, and spiral arms rich in young stars, gas and dust. Subtypes (Sa, Sb, Sc, SBa, SBb, SBc) reflect the looseness of spiral arms, bulge-to-disc ratio, and bar presence (Turner et al., 2016). These galaxies are key laboratories for studying star-formation and secular evolution.
Lenticular (S0) galaxies provide an intermediate class: they have a disc and bulge, but lack prominent spiral arms and active star formation, and are often considered as evolving from spirals via gas removal or quenching (Turner et al., 2016).
Irregular galaxies have no defined structure, often chaotic, undergoing active star formation and shaped by interactions or environmental disruption. They challenge simple classification schemes and reflect dynamic processes (Buta, 2010).
Peculiar/merging galaxies exhibit distortions, tidal tails, warped discs, or other features due to collisions or interactions; they provide insight into galaxy assembly and morphological transformation.
Dwarf galaxies (elliptical, irregular, or compact) are low-mass systems that serve as building blocks in hierarchical structure formation and hold clues to dark matter and cosmic evolution (Buta, 2010).
Active galaxies are characterised by energetic nuclei powered by supermassive black holes (e.g., Seyfert, radio galaxies, quasars). Their morphological context and host classification reveal co-evolution of galaxies and black holes.

The classification of these types remains astrophysically meaningful: morphology correlates with star-formation rate, stellar age, gas fraction and environment (Conselice, 2006; Buta, 2010). Thus, improving classification accuracy and scale directly supports galaxy-evolution studies.

AI and Automated Morphology Classification

The challenge facing modern astronomy is that large sky surveys (e.g., SDSS, DES, LSST) generate millions of galaxy images and multi-wavelength data. Manual classification becomes untenable. AI techniques—particularly supervised machine learning, convolutional neural networks (CNNs), transfer learning, semi-supervised learning and generative models—have emerged as powerful tools (Ball & Brunner, 2010; Dieleman et al., 2015).

Initial approaches applied decision-tree, random-forest and fuzzy logic algorithms to photometric parameters (Gauci et al., 2010). Yet image-based methods using CNNs yield substantially higher accuracy: for instance, a CNN trained on the Galaxy Zoo 2 dataset with ~28,790 images achieved over 95% classification accuracy into five morphological classes (Dai & Tong, 2018). More recently, CNNs applied to binary classification (elliptical vs spiral) reached ~99% accuracy using DES data (Domínguez Sánchez et al., 2020). Transfer-learning approaches (using pretrained networks such as AlexNet) and new architectures (e.g., Convolutional vision Transformer, CvT) further push accuracy and robustness (Becker et al., 2023; A&A 2024). Semi-supervised methods such as dynamic threshold alignment (DTA) address the challenge of limited labelled data and class imbalance (Jiang et al., 2023). The breadth of methods is reviewed in recent surveys (Barchi et al., 2019; Luo et al., 2025).

Application to Specific Galaxy Types

Elliptical galaxies: AI methods differentiate ellipticals from spirals with high accuracy using morphological images and/or photometric parameters (Turner et al., 2016). CNNs excel at identifying the smooth light profiles and lack of disc structure (Domínguez Sánchez et al., 2020).
Spiral galaxies: AI models classify normal vs barred spirals (Sa, Sb, Sc vs SBa, SBb, SBc) and quantify features like arms, bars and bulge strength (Zhang et al., 2019). Machine learning on multi-wavelength images enables estimation of star-formation rates and mapping of young stellar populations in arms.
Lenticular galaxies: The intermediate morphology of S0 galaxies poses classification challenges; however, supervised ML methods (e.g., SVM, random forest) on photometric features (color indices, concentration) achieved binary early/late type accuracy ~95% in SDSS DR9 (Turner et al., 2016). Image-based networks refine this further by distinguishing S0 from E and S types.
Irregular and peculiar galaxies: Anomaly detection and unsupervised learning methods identify disturbed systems, merger candidates and irregular morphologies (Hocking et al., 2017). Generative DL models simulate morphological diversity in merging systems (Lucie-Smith et al., 2018).
Dwarf galaxies: Because of low surface brightness and small size, detection and classification of dwarfs benefit from ML pipelines trained on deep imaging and tailored feature extraction (Pasquet et al., 2019).
Active galaxies: Time-series and image-based DL methods track variability in AGNs and classify host morphology, aiding in the study of galaxy–black hole co-evolution (Kozłowski, 2016).
Overall, AI allows large-scale morphological catalogues: for example, a DL-based catalogue of ~164,000 galaxies in the S-PLUS DR3 achieved ~98.5% precision distinguishing late vs early-type morphology (Bom et al., 2023).

Benefits, Challenges and Future Prospects

AI-driven morphology studies yield several benefits:

• Scalability: ability to process millions of galaxies rapidly, far beyond manual visual classification.
• Depth: extraction of subtle structural features (bars, rings, warps) and estimation of physical properties (star-formation rate, bulge/disc ratio).
• Novel discovery: anomaly detection identifies rare or peculiar systems (e.g., mergers, tidal features).
• Integration: multi-wavelength, time-series and simulation data can be combined to model morphological evolution.

However, challenges remain:

• Label bias and training‐set quality: supervised methods require large, reliable labelled data and may inherit biases from citizen science or expert labels (Turner et al., 2016).
• Interpretability: DL models often act as “black-boxes”; linking morphological classifications to astrophysical processes demands careful validation (Weigert et al., 2023).
• Imbalanced classes & rare types: dwarf galaxies, mergers, low‐surface-brightness systems remain under-represented; semi-supervised and generative methods can help (Jiang et al., 2023).
• Generalisation across surveys: variations in depth, resolution, band-passes and redshift require robust architectures and domain adaptation (Becker et al., 2023).
• Physical interpretation and simulation integration: classification must be connected with galaxy‐evolution models and simulation outputs to yield insight, not just labels.

Future directions include applying transformer architectures for morphological tasks (A&A 2024), incorporating temporal/variability data for active galaxies, coupling classification with simulation‐based synthetic data generation, and implementing human–AI hybrid workflows (Beck et al., 2018). As next‐generation surveys such as Legacy Survey of Space and Time (LSST) and Euclid commence, the need for scalable, accurate AI‐enabled classification becomes critical.

---

Conclusion

Morphological classification of galaxies remains a cornerstone of extragalactic astrophysics, linking structure to formation and evolution. The diversity of galaxy types—elliptical, spiral (normal and barred), lenticular, irregular, peculiar/merger, dwarf and active—encapsulates a wide range of physical parameters and evolutionary states. The advent of AI, particularly deep learning and semi-supervised approaches, has transformed how astronomers study morphology: enabling high accuracy classification at scale, discovery of subtle morphological features, and efficient handling of large complex data sets. Despite challenges in bias, interpretability and class imbalance, the integration of AI and astrophysical interpretation holds tremendous promise. For the coming era of massive surveys, the synergy of AI techniques with morphological theory and simulation will be essential to unlock new understanding of galaxy formation and the evolving Universe.

---

References

Ball, N. M., & Brunner, R. J. (2010). Data mining and machine learning in astronomy. International Journal of Modern Physics D, 19(07), 1049–1106. https://doi.org/10.1142/S0218271810017160
Barchi, P. H., de Carvalho, R. R., Rosa, R., Sautter, T., Soares-Santos, M., Marques, B. A. Z., … Moura, T. C. (2019). Machine and Deep Learning Applied to Galaxy Morphology – A Comparative Study. arXiv. https://arxiv.org/abs/1901.07047
Becker, M. R., Scarlata, C., Fortson, L. F., Simmons, B. D., Galloway, M. A., Willett, K. W., … Wright, D. (2018). Integrating human and machine intelligence in galaxy morphology classification tasks. arXiv. https://arXiv.org/abs/1802.08713
Bom, C. R., Cortesi, A., Ribeiro, U., Dias, L. O., Kelkar, K., Smith Castelli, A. V., … Abramo, L. R. (2023). An Extended Catalogue of Galaxy Morphology Using Deep Learning in S-PLUS DR3. Monthly Notices of the Royal Astronomical Society, 528(3), 4188–4205. https://doi.org/10.1093/mnras/sty/7492270
Buta, R. J. (2010). Galaxy morphology. In Planets, Stars, and Stellar Systems, Volume 6 (pp. 5–81). Springer. https://doi.org/10.1007/978-1-4020-9218-9_2
Conselice, C. J. (2006). The fundamental properties of galaxies and a new galaxy classification system. Monthly Notices of the Royal Astronomical Society, 373(4), 1389–1409. https://doi.org/10.1111/j.1365-2966.2006.11122.x
Dai, J.-M., & Tong, J. (2018). Galaxy Morphology Classification with Deep Convolutional Neural Networks. arXiv. https://arXiv.org/abs/1807.10406
Dieleman, S., Willett, K. W., & Dambre, J. (2015). Rotation-invariant convolutional neural networks for galaxy morphology prediction. Monthly Notices of the Royal Astronomical Society, 450(2), 1441–1459. https://doi.org/10.1093/mnras/stv632
Domínguez Sánchez, H., Huertas-Company, M., Bernardi, M., Tuccillo, D., & Fischer, J. L. (2020). Optimizing Automatic Morphological Classification of Galaxies with Deep Learning. Monthly Notices of the Royal Astronomical Society, 493(3), 4209–4229. https://doi.org/10.1093/mnras/staa074
Gauci, A., Zarb Adami, K., & Abela, J. (2010). Machine Learning for Galaxy Morphology Classification. arXiv. https://arXiv.org/abs/1005.0390
Hocking, A., Geach, J. E., Sun, Y., & Davey, N. (2017). An Automatic Taxonomy of Galaxy Morphology Using Unsupervised Machine Learning. arXiv. https://arXiv.org/abs/1709.05834
Jiang, J., Zhang, J., Li, X., Li, H., & Du, P. (2023). Galaxy Morphology Classification Using a Semi-supervised Learning Algorithm Based on Dynamic Threshold. Research in Astronomy and Astrophysics, 23(11), 115019. https://doi.org/10.1088/1674-4527/acf610
Kozłowski, S. (2016). Quantifying quasar variability as a damped random walk. The Astrophysical Journal, 826(2), 118. https://doi.org/10.3847/0004-637X/826/2/118
Lotz, J. M., Primack, J., & Madau, P. (2003). A New Non-Parametric Approach to Galaxy Morphological Classification. arXiv. https://arXiv.org/abs/astro-ph/0311352
Lucie-Smith, L., et al. (2018). Generating Realistic Galaxies with Deep Generative Models. Monthly Notices of the Royal Astronomical Society, 480(2), 2743–2756. https://doi.org/10.1093/mnras/sty2014
Pasquet, J., Trémolet, Y., Massonet, F., & Colin, J. (2019). Deep Learning for Photometric Redshift Estimation: Applications to SDSS, DES, and LSST. Astronomy & Astrophysics, 621, A26. https://doi.org/10.1051/0004-6361/201833408
Turner, R. J., et al. (2016). Machine Learning Technique for Morphological Classification of Galaxies. Astronomy & Astrophysics, 592, A60. https://doi.org/10.1051/0004-6361/201628633
van den Berg, S. (1998). Galaxy Morphology and Classification. Cambridge University Press.
Weigert, C., et al. (2023). Machine Learning and Galaxy Morphology: For What Purpose? arXiv. https://arXiv.org/abs/2306.02626
Zhang, Z., et al. (2019). Machine Learning for Star Formation Rates in Galaxies. The Astrophysical Journal, 876(2), 110. https://doi.org/10.3847/1538-4357/ab11f3