Tucker Adeyemi | 2026 I.S. Symposium
Name: Tucker Adeyemi
Title: Creating a CNN-Based App for Classifying Galaxies
Major: Computer Science
惭颈苍辞谤:听Physics
Advisor: Sofia Visa
Deep learning models, specifically Convolutional Neural Networks (CNNs), have revolutionized the field of computer vision, demonstrating exceptional proficiency in complex image recognition tasks like facial recognition. However, a significant gap exists in the application of these technologies to the physical sciences. Most accessible pretrained models are optimized for terrestrial objects and fail to account for the unique structural nuances of astronomy-based imagery. As someone deeply passionate about both astronomy and programming, I found this project to be an exciting opportunity to bridge the two fields by incorporating machine intelligence into the study of the cosmos. This research involved training a specialized CNN architecture to categorize images of galaxies into five distinct morphological classes. Utilizing a dataset of over 60,000 images from the Galaxy Zoo 2 project, I was able to develop a functional web application that allows users to upload galaxy images for real-time classification. The resulting model achieved an accuracy rate ranging from 70% to 90%, proving that deep learning is a highly feasible and efficient method for large-scale celestial classification. The most exciting conclusion drawn from this work is that even a relatively compact model can effectively learn complex hierarchical features of distant galaxies. However, the research also highlighted limitations, particularly in accurately detecting irregular galaxies. Future work will focus on incorporating larger, more varied datasets to improve the model’s versatility. I see potential in expanding the classification system to include more granular categories, such as barred spiral galaxies, to further refine our automated understanding of the universe’s structure.
Posted in Symposium 2026 on May 1, 2026.
