Galaxy Morphological Image Classification using ResNet

  • Siddhartha Banerjee Ramakrishna Mission Residential College (Autonomous), Narendrapur, Kolkata, West Bengal, India https://orcid.org/0000-0001-5353-4522
  • Bibek Ranjan Ghosh Ramakrishna Mission Residential College (Autonomous), Narendrapur, Kolkata, West Bengal, India
  • Ayan Gangapadhyay Ramakrishna Mission Residential College (Autonomous), Narendrapur, Kolkata, West Bengal, India
  • Himadri Sankar Chatterjee Ramakrishna Mission Residential College (Autonomous), Narendrapur, Kolkata, West Bengal, India https://orcid.org/0000-0001-6110-6229
Keywords: Convolutional Neural Network, Residual Networks, ResNet-18, Galaxy Zoo Data set, Galaxy Morphological Classification

Abstract

     Machine learning-based techniques are used widely for the classification of images into various categories. The advancement of Convolutional Neural Network (CNN) affects the field of computer vision on a large scale. It has been applied to classify and localize objects in images. Among the fields of applications of CNN, it has been applied to understand huge unstructured astronomical data being collected every second. Galaxies have diverse and complex shapes and their morphology carries fundamental information about the whole universe. Studying these galaxies has been a tremendous task for the researchers around the world. Researchers have already applied some basic CNN models to predict the morphological classes of the galaxies. In this paper, a residual network (ResNet) model is applied for this purpose. The proposed methodology classified the galaxies depending on their shape into 37 different classes. The performance of the methodology was evaluated using the data set provided by Kaggle. In this data set, 61,578 galaxy images are given, which are classified by human eye. The model achieved nearly 98% accuracy.

Published
2021-10-30
How to Cite
Siddhartha Banerjee, Bibek Ranjan Ghosh, Ayan Gangapadhyay, & Himadri Sankar Chatterjee. (2021). Galaxy Morphological Image Classification using ResNet. Iraqi Journal of Science, 62(10), 3690-3696. https://doi.org/10.24996/ijs.2021.62.10.27
Section
Computer Science