Celestial Image Classification Using Attention And Boosting Mechanism

Title

Celestial Image Classification Using Attention And Boosting Mechanism

Creator

Varghese, Nisha; Justin, Jesvin K; Kezya Steffyn, S.

Description

Astronomical image classification is vital in the comprehension of celestial objects, but deep learning models are severely challenged by the lack of labeled datasets. The novelty of the study is two-fold - the development of the dataset and a hybrid learning method that combines both transformer-based feature extraction and gradient-boosted decision trees to improve classification performance for celestial image classification. This study is a comparison of CNNs, transformers, and hybrid models in nebulae, galaxy, and star cluster classification using the dataset collected from the Hubble Space Telescope image archive. Through progressive data augmentation, the dataset was augmented from 603 images to 4,500 high-diversity training samples to enhance model generalization. This research explores various architectures, including ResNet-50, DenseNet-121, EfficientNetV2-S, DeiT (Data-Efficient Image Transformer), and hybrid models like DeiT-RF (Data-Efficient Image Transformer - Random Forest) and DeiT-XGBoost (DXg). DXg brings a novel fusion mechanism in which DeiT learns high-level spatial representations, adaptive dimensionality reduction fine-tunes feature selection, and XGBoost best classifies celestial objects. Such a unique combination of transformers and gradient boosting improves interpretability without sacrificing state-of-the-art performance. 2025 IEEE.

Source

2025 IEEE Space, Aerospace and Defence Conference, SPACE 2025;

Date

01-01-2025

Publisher

Institute of Electrical and Electronics Engineers Inc.

Subject

Astronomical Image Classification; Convolutional Neural Networks (CNNs); Data Augmentation; Deep Learning; Dimensionality Reduction; Ensemble Learning; Hybrid Models; Transformers; XGBoost

Coverage

Varghese N., Christ University, Department of Computer Science, Karnataka, Bangalore, 560029, India; Justin J.K., Christ University, Department of Computer Science, Karnataka, Bangalore, 560029, India; Kezya Steffyn S., Christ University, Department of Computer Science, Karnataka, Bangalore, 560029, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISBN: 979-833151552-2;

Format

online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1109/SPACE65882.2025.11170481

https://www.scopus.com/pages/publications/105018457544?origin=resultslist