Globular Cluster Detection in M33 Using Multiple Views Representation Learning

Taned Singlor, Phonphrm Thawatdamrongkit, Prapaporn Techa-angkoon, Chutipong Suwannajak, Jakramate Bootkrajang

Abstract: Globular clusters (GC) are crucial for understanding galaxy formation and evolution. However, identifying them in large imagery datasets is a time-consuming task. This prompts the development of an automated GC detection algorithm. Although GC detection is fundamentally an object detection problem, the state-of-the-art object detection algorithms are unable to produce accurate results. Motivated by how GCs are identified by astronomers, we propose a deep neural network that fuses multiple views of raw imaging data and learns a better representation of the input image. The proposed network is then combined with YOLO object detection algorithm resulting in YOLO for Globular Cluster detection (YOLO-GC) model. Experimental results based on a real catalog of GCs in the M33 Galaxy showed that the proposed multi-view representation learning technique helps improve detection performance.