Fine-grained Morphological Study of Galaxy Pairs: A Human-in-the-loop Self-supervised Classification Framework

Abstract Galaxy pairs, as key probes for studying galaxy formation and evolution, provide a unique window for observing galaxy interactions, including the associated exchange of material and related galaxy formation activities. However, the analysis of complex evolutionary scenarios of galaxy pairs using photometric image projections faces significant challenges due to factors such as significant intraclass morphological differences, interference from background noise, interclass feature overlap, and inconsistent observational conditions. This study proposes a human-in-the-loop, self-supervised framework for fine-grained morphological classification of galaxy pairs based on multilevel contrastive clustering, and constructs a seven-category system comprising close galaxy pairs, projection-separated galaxy pairs, edge-on galaxy pairs, irregular galaxy pairs, large-scale spiral galaxy pairs, tidal tail galaxy pairs, and outlier galaxy pairs. Through a systematic analysis of the distribution of key parameters including redshift, line-of-sight velocity difference, mass, and star formation rate across various types of galaxy pairs, this study reveals that distinct types of galaxy pairs exhibit significant disparities in redshift correlation, mass composition, and star formation activity. Comparison with existing classification methods further demonstrates that this fine-grained morphological classification framework exhibits superior discriminative ability in distinguishing galaxy pairs with different morphologies.