Image-Based Constellation Recognition Using Deep Learning on Consumer-Grade Sky Imagery

Bhakti Israni

Abstract: Understanding the night sky has long been a source of scientific curiosity, yet for most people it remains abstract and inaccessible. Although several applications claim to identify stars and constellations, their outputs are typically derived from positional sensor data rather than from direct analysis of the visual sky itself. As a result, these tools do not truly interpret what is visible in an image, nor do they address the broader computer-vision challenge of recognizing astronomical patterns under real-world conditions. This paper investigates whether modern deep-learning techniques can be used to directly analyze consumer-grade night-sky images captured using smartphones. We propose a vision-based constellation recognition system trained on a combination of synthetic sky renderings and real, low-light photographs. Multiple architectures-including convolutional neural networks, Vision Transformers, and a hybrid CNN-Transformer model-are evaluated for accuracy, robustness, and computational efficiency. The best-performing model is integrated into a web-based application to demonstrate real-world feasibility. By focusing on image-based interpretation rather than sensor inference, this work contributes to both applied computer vision and accessible astronomy education.

Keywords: Deep Learning, Hybrid CNN-Transformer Models, Constellation Recognition, Low-Light Image Recognition, Night-Sky Image Analysis

How to Cite?: Bhakti Israni, "Image-Based Constellation Recognition Using Deep Learning on Consumer-Grade Sky Imagery", Volume 14 Issue 12, December 2025, International Journal of Science and Research (IJSR), Pages: 1632-1636, https://www.ijsr.net/getabstract.php?paperid=SR251220164234, DOI: https://dx.doi.org/10.21275/SR251220164234