Optimization Residual Network for Spiral Galaxy Spin Direction Classification Hafiz Indra Arwinata (a*), Sultan Hadi Kusuma (a), Anton Timur Jaelani (b)
a) Astronomy Study Program, Faculty Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesa 10, Bandung 40132, Indonesia
* 20324002[at]mahasiswa.itb.ac.id
b) Astronomy Research Group and Bosscha Observatory, Faculty Mathematics and Natural Sciences, Institut Teknologi Bandung, Jl. Ganesa 10, Bandung 40132, Indonesia
Abstract
Spiral galaxies are identified as Z-spiral or S-spiral based on their spin directions. This spin direction relates to the galaxy formation. Hence, the distribution between Z-spiral and S-spiral galaxies offers significant insights into galaxy formation and evolution. Recently, large survey programs, such as the Dark Energy Spectroscopic Instrument (DESI) Legacy Imaging Survey and Hyper Suprime-Cam Subaru Strategic Program (HSC SSP), have been providing huge amounts of high-quality data on galaxies. Hence, the galaxy morphology classification cannot be done conventionally by visual inspection, and machine learning plays a prominent role in this classification process. This study investigates the use of Residual Network (ResNet) for classifying spiral galaxies using data from the DESI and HSC SSP. We use the ResNet-34 model by using a data augmentation procedure based on several criteria during the training process. This architecture model achieves a robust accuracy of up to 90% in the data testing process. This performance shows great potential for large-scale galaxy surveys, where conventional methods of visual inspection are ineffective.
