Constellation identification method using point set data

Abstract

The rapid development of machine learning has made it possible to recognize images with high accuracy across a wide range of tasks. However, constellations found through astrophotography are difficult to identify because of a lack of datasets necessary for training artificial intelligence models. In addition, the fact that the appearances of stars vary significantly depending on the photographer and the equipment used, even in photographs of the same constellation, makes it difficult to identify constellations. Owing to these issues, few studies have explored constellation identification. This study attempted to address the identification of constellations, which has not been well studied, from a unique angle. Specifically, we propose a method for identifying constellations found through astrophotography by viewing constellations as sets of points using datasets created based on a previous study. We confirmed through numerical experiments that the proposed method provides a higher identification rate than the case without the proposed method. We also gained perspectives on how to improve the identification rate.