Image Classification Algorithm for Determining the Light Curve Morphologies of ASAS-SN Eclipsing Binaries

TL;DR

This paper introduces ASEBCLASS, a deep learning-based image classification algorithm that accurately categorizes eclipsing binary star light curves into three classes with 92% accuracy, aiding astronomical analysis.

Contribution

The paper develops a convolutional neural network model and a Python code for classifying eclipsing binary light curve morphologies from images, achieving high accuracy.

Findings

Achieved 92% classification accuracy.

Successfully distinguished three binary star classes.

Provided a Python tool for automated light curve classification.

Abstract

We present a classification of the light curve morphologies of eclipsing binary systems observed by ASAS-SN based on their light curve images. The data of 16500 eclipsing systems having three different classes (detached Algol type, $\beta$ Lyr type, and W UMa type) are collected to construct their light curves. A deep learning algorithm containing the convolutional neural networks is employed on the images to achieve a satisfying classification. A code called ASEBCLASS is written in Python language, and it uses the TensorFlow platform through Keras API to employ the mathematical libraries in the training phase of the model. The architecture consists of four groups of convolutional, activation, maximum pooling layers together with the additional fully connected layers. The results show that our algorithm estimates the morphological class of an external input image data with an accuracy value of 92%.