Date of Award
Spring 2026
Document Type
Open Access Thesis
Department
Statistics
First Advisor
Yuexuan Wu
Abstract
Galaxy morphology classification is an essential task for understanding galaxy evolution, but modern high-redshift surveys such as JADES produce a large number of unlabeled images that cannot be handled with traditional manual or supervised learning methods. This work explores a semi-supervised transfer learning framework that leverages labeled data from Galaxy Zoo 2 to classify morphologies in James Webb Space Telescope JADES observations. A pretrained Convolutional Vision Transformer is adapted to the new imaging domain and refined using pseudo-labeling on unlabeled data. Experiments simulating JADES imaging conditions show that reduced resolution and noise significantly impact the separability of certain morphology classes, particularly edge-on and cigar-shaped smooth galaxies. Despite these challenges, the adapted model successfully transfers morphological knowledge across domains, achieving substantial accuracy improvements over baseline methods that do not account for domain shift. These results demonstrate the potential of transfer learning for galaxy morphology classification while exposing the limitations imposed by domain shift.
Rights
© 2026, Emily Miller Christen
Recommended Citation
Christen, E. M.(2026). Semi-Supervised Galaxy Morphology Classification Using Convolutional Vision Transformers. (Master's thesis). Retrieved from https://scholarcommons.sc.edu/etd/8888