A Deep Learning Approach for Identification and Classification of Exoplanets

Abstract

More than a million stars have been observed in the last decade for transiting planets. The manual interpretation of candidates for exoplanets is labor-intensive and susceptible to human mistake, the consequences of which are difficult to measure. A neural network, rather than the more traditional ways of discovering exoplanet candidates, is being used in large scale planetary search operations. Neural networks, often known as ”deep learning” or ”deep nets,” are designed to offer a computer understanding of a given problem by teaching it to identify patterns. As a result, Earth-sized planets orbiting Sunlike stars have eluded NASA’s Kepler Space Telescope, which was built to find out how common they are. Individual candidates for planets will need to be automatically and precisely assessed for their likelihood of becoming planets, even at low signal-to-noise ratios. Deep learning, a family of machine learning methods that has recently become state-of-the-art in a wide range of problems, is used in this paper to classify probable planet signal. Deep convolutional neural networks are used to identify whether an exoplanet transiting the star is real or a false positive caused by an astrophysical or instrumental event. There is a huge research gap between the identification of such exoplanets at such highest level of complex artificial intelligence model. The major gap analysis is the time required to identify the planet is an exoplanet or not an exoplanet. The proposed model also adds the state of art extension algorithm which is capable of classifying the exoplanets into standard types of the planets. The goal of this project is to analyze data from the Kepler telescope and create a state of art model that can identify exoplanets and classify them into different types of planets time efficiently and with high accuracy.