Detection of Southern Hemisphere Constellations Using Convolutional Neural Networks

Constellations allow the identification of most stars and celestial objects visible in the night sky at a glance, without the use of a telescope. However, due to the large number of constellations, this task can be overwhelming for astronomy beginners. To perform this identification, many rely on mobile applications that depend on Internet connectivity and Global Positioning System. Unfortunately, these applications only provide estimates based on geolocation and do not guarantee an accurate visual representation. For this reason, this paper proposes the identification of constellations using Convolutional Neural Networks. The purpose of this research is to detect constellations with greater accuracy from photographs of any size, regardless of the availability of an Internet connection. For this, a convolutional neural network model called You Only Look Once was used. This neural network is widely known for its accuracy in object detection and is ideal for pattern recognition, such as constellations. In this work, different versions of this neural network were used to detect the 21 most representative constellations of the southern hemisphere. The results obtained reveal that all models exhibit outstanding performance, with high precision and recall values, resulting in F1-<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">scores</i> of 0.991 and higher.