Using Deep Learning with Phase Folded Light Curves to Detect Exoplanets

Exoplanets are planets that orbit stars other than our sun, and one way of detecting them is by the transit method. The transit method consists of plotting the apparent brightness of the exoplanet’s host star against a comparison star to make a light curve, in which periodic dips in brightness can correspond to the planet coming in front of the star. However, detecting if the dip in brightness comes from the existence of an exoplanet, and not an extraneous systematic, remains a challenging problem, due to small dips in brightness and noisy observations. Deep learning techniques have shown promise in many areas such as exoplanet detection from light curves, Nbody analysis, and even atmospheric characterization. In this study, we extend the work of transit detection by utilizing phase folded light curves to train a convolutional neural network to predict the existence of an exoplanet. We propose a scheme to generate synthetic data mimicking TESS observations from a single sector where the positive training samples are randomly generated transit data phase folded at the correct period, and the negative training samples are folded at a random incorrect period. Our model trained on this data can be applied to both real and synthetic data to make an assessment as to whether the data contains an exoplanet transit. Furthermore, we will compare our detection algorithm to the Transit Least-Squares (TLS) algorithm and observe the effects of signal aliasing.