A robust morphological classification method for galaxies using dual-encoding contrastive learning and multi-clustering voting on JWST/NIRCam images

TL;DR

This work tackles the challenge of scalable, accurate galaxy morphology classification in large JWST COSMOS-Web datasets. It advances a two-step framework (USmorph) by integrating CAE denoising, APCT rotational normalization, and a dual-encoder contrastive learning scheme using ConvNeXt and ViT, followed by PCA for dimensionality reduction. The unsupervised clustering (bagging across three algorithms) yields 17,326 reliably labeled galaxies, which train a GoogLeNet classifier that labels the remaining $\sim$28,850 galaxies with a final accuracy of about $94.6\%$. Morphological parameter analyses (parametric and nonparametric) validate the classifications, demonstrating consistent trends (e.g., $n$, $r_e$, $G$, $M_{20}$, $C$, $\Psi$, and MID metrics) across SPH, ETD, LTD, and IRR types, and underscoring the framework’s potential for upcoming large-sky surveys such as those from the Chinese Space Station Telescope.

Abstract

The two-step galaxy morphology classification framework {\tt USmorph} successfully combines unsupervised machine learning (UML) with supervised machine learning (SML) methods. To enhance the UML step, we employed a dual-encoder architecture (ConvNeXt and ViT) to effectively encode images, contrastive learning to accurately extract features, and principal component analysis to efficiently reduce dimensionality. Based on this improved framework, a sample of 46,176 galaxies at $0<z<4.2$, selected in the COSMOS-Web field, is classified into five types using the JWST near-infrared images: 33\% spherical (SPH), 25\% early-type disk (ETD), 25\% late-type disk (LTD), 7\% irregular (IRR), and 10\% unclassified (UNC) galaxies. We also performed parametric (S{é}rsic index, $n$,and effective radius, $r_{\rm e}$) and nonparametric measurements (Gini coefficient, $G$, the second-order moment of light, $M_{\rm 20}$, concentration, $C$, multiplicity, $Ψ$, and three other parameters from the MID statistics) for massive galaxies ($M_*>10^9 M_\odot$) to verify the validity of our galaxy morphological classification system. The analysis of morphological parameters is consistent with our classification system: SPH and ETD galaxies with higher $n$, $G$, and $C$ tend to be more bulge-dominated and more compact compared with other types of galaxies. This demonstrates the reliability of this classification system, which will be useful for a forthcoming large-sky survey from the Chinese Space Station Telescope.

Figures (11)

• Figure 1: Relationship between stellar mass and redshift. The solid red line shows the stellar-mass completeness, described by $M_{\rm comp}(z)/M_{\odot} = -1.51 \times 10^6 (1 + z) + 6.81 \times 10^7 (1 + z)^2$.
• Figure 2: Schematic of the improved galaxy morphology classification system. Panel (a) shows the flow of the galaxy morphology classification system. Panel (b) illustrates feature extraction using ConvNeXt and ViT models to encode the data and extract key features. Encoding results from the same data instances are used to construct positive sample pairs, while the remaining encoding from the ViT model form negative sample sets. The contrastive loss (CL) function is applied to minimize the loss and extract important features. Finally, dimensionality reduction is performed using PCA on the encoded results after contrastive learning.
• Figure 3: Six sets of images demonstrating image preprocessing steps. In each set, the left, center, and right panels show the original images in the NIR band, post-CAE-based denoised images, and images after polar coordinate expansion, respectively.
• Figure 4: Random selection of 100 images for each of five galaxy types, identified from 20 groups through visual inspection. The clustering results show small morphological differences within groups and large differences between groups, enabling visual classification of each component into SPH, ETD, LTD, IRR, and UNC types.
• Figure 5: Galaxy morphology classification performance: recall, precision, and general confusion matrix analysis. Left and middle panels represent the recall and precision of the GoogLeNet model, with both exceeding 94.6%, which indicates that the GoogLeNet model can effectively distinguish between different types of galaxies. The right panel represents the general confusion matrix.