SIMBAD references

Context. Current and near-future sky survey programmes, such as the Legacy Survey of Space and Time (LSST) of the Vera C. Rubin Observatory, will produce vast amounts of data that will need new techniques to process them on reasonable timescales. Machine-learning methods and properly trained neural networks proved to be efficient, fast, and reliable in performing a variety of tasks, such as the classification of variable star light curves. Since LSST survey full sky data are not yet available (only the data preview 1 from various sky segments), we tested our methods that are to be used on real LSST data on proxy datasets.
Aims. In this project, we used data obtained by the Zwicky Transient Facility to develop and test a neural-network-based multiband classification algorithm for classifying periodic variable stars (i.e. pulsating variable stars and eclipsing binaries). The aim is to use the algorithm on LSST data when they become available.
Methods. Phase-folded light-curve images and period information were used from five different variable star types: Classical and Type II Cepheids, δ Scuti stars, eclipsing binaries, and RR Lyrae stars. The data were taken from the 17th data release of ZTF, from which we used two passbands, g and r, in this project. The periods were calculated from the raw data, and this information was used as an additional numerical input in the neural network. For the training and testing process, a supervised machine-learning method was created. The neural network contained convolutional neural networks concatenated with fully connected layers.
Results. During the training-validation process, the training accuracy reached 99% and the validation accuracy peaked at 95.6%. In the test classification phase, three variable star types out of the five classes were classified with an accuracy of about 99%. The other two had a very high accuracy as well, 89.6% and 93.6%.
Conclusions. Our results showed that by using phase-folded light curves from multiple passbands and the periods as numerical data inputs, we are able to train a neural network with outstanding accuracy.