Vision-Language Model Ensembles Achieve Human-Expert Accuracy for Galaxy Merger Classification

Abstract

We present a proof-of-concept study demonstrating that an ensemble of Vision--Language Models (VLMs) combined using a Bayesian statistical framework can classify galaxy merger morphologies with accuracy comparable to trained human experts. We deploy 15 VLM classifier configurations, spanning four model architectures (Gemma-4 E2B, Gemma-4 E4B, Qwen2.5-VL, and Qwen3-VL) tested with up to four prompt engineering strategies each. We evaluate their performance against a truth-known sample of 41 VELA+SUNRISE mock galaxy images from Lambrides et al. 2021. The VLM ensemble achieves 83.3\% accuracy on confident classifications (merger probability $p_{\rm M} \ge 0.8$ or $p_{\rm M} \le 0.2$), with 5 misclassified galaxies. The ensemble recovers the population merger fraction to within $0.66\sigma$ of the truth ($f_{\rm M} = 0.52 \pm 0.09$ vs.\ true value of 0.585). Bayesian weighting improves overall accuracy by 17.1 percentage points over simple majority voting, with sensitivity improving by 29.2 percentage points. The VLM ensemble produces 5 misclassified galaxies (2 false positives, 3 false negatives), comparable to the 6 misclassifications (5 false positives, 1 false negative) reported for human classifiers by Lambrides et al. 2021. The apparent differences in error profiles are not statistically significant given the sample size of 41 galaxies. VLMs also produce more moderate per-galaxy merger probability distributions (27\% uncertain) than the more polarized human distributions (15\% uncertain), though this difference is also consistent with statistical fluctuation. These results establish VLMs as scalable, reproducible alternatives to human classifiers within a Bayesian probabilistic merger-fraction framework, with direct applications to large galaxy samples from current and future surveys.