L:Deep Learning Approaches for Exoplanet Detection using the Transit Method on Light Curve Data

Exoplanet detection has appeared as one of the most interesting edges in modern astronomy, giving meaningful insights about planetary systems that are beyond our imagination. With over 5,000 confirmed exoplanets discovered to date, there are many detection techniques that includes radial velocity, direct imaging, and gravitational microlensing have remarkable contribution to this field. Even though there are several methods to detect the exoplanets, transit method is exceptional for its ability to provide information about a planet size, orbital period, and potential habitability. The Transiting Exoplanet Survey Satellite telescope has played a vital role in the field of astronomy by collecting huge amount of light curve data from millions of stars, understanding that life exists beyond earth. Separating true planetary transits from noisy and other stellar variations within this data remains a challenging and resource-intensive task. To enhance the detection accuracy deep learning models such as Feedforward Neural Networks , Bayesian Neural Networks , and Recurrent Neural Networks are trained to understand the patterns in light curve data. These models have detection accuracies of 95%, 91%, and 96%, respectively. By leveraging the uniqueness of each model FNN for structured data, BNN for uncertainty estimation, and RNN work better with time-series data, these methods work together effectively to classify the planetary transit events. This research aims to enhance the exoplanet discovery, enhance classification accuracy, and reduce computational costs when compare to object detection models and reduce time consumption.