DEEP LEARNING EXOPLANETS DETECTION BY COMBINING REAL AND SYNTHETIC DATA

Abstract: Our work combines real observation data with synthetic data in exoplanet detection by deep learning. We collate datasets from space-based telescopes, like Kepler and TESS, from ground-based observatories for light curves, and spectral data. To further enrich the dataset, we produce synthetic data simulating a collection of astrophysical scenarios. Convolutional and recurrent neural networks enable model robustness and generalization. Accuracy and reliability of exoplanet detection will increase with training using the total dataset. Such integration will not only extend the scope of the training dataset to probe a far greater variety of astrophysical conditions but also speed up the discovery and characterization of exoplanets.