A Comprehensive Investigation of Attention-Based CNNs for Galaxy Morphology Classification

Abstract

Galaxy morphology classification is a fundamental yet challenging task in astronomical research, as it plays a crucial role in understanding the universe's structure and evolution. The complexity and diversity of galaxy shapes, combined with the large volume of astronomical data, necessitate efficient and accurate automated methods for classification. This paper provides a comprehensive review of attention-based convolutional neural networks (CNNs) for galaxy classification, highlighting their potential to overcome limitations of conventional approaches. This paper systematically surveys three advanced methods: attention-gating, multi-branch attention networks, and dynamic multiscale attention networks, detailing their architectures, mechanisms, and performance gains. These methods enhance feature focus, multi-scale fusion, and interpretability while reducing computational costs. The analysis confirms that attention-based CNNs significantly improve classification accuracy and robustness, making them highly valuable for large-scale astronomical surveys. This review offers valuable insights into current advancements and future directions, contributing to the development of more reliable and efficient galaxy classification systems.