A machine learning technique with two-dimension convolutional neural network is proposed for detecting exoplanet transits. To test this new method, five different types of deep learning models with or without folding are constructed and studied. The light curves of the Kepler Data Release 25 are employed as the input of these models. The accuracy, reliability, and completeness are determined and their performances are compared. These results indicate that a combination of two-dimension convolutional neural network with folding would be an excellent choice for the future transit analysis. successfully discovered new transits from Kepler light curves. Moreover, without Box-Least-Square algorithm, Pearson et al. (2018) suggested to directly use deep learning to examine the folded Kepler light curves to search for transits. Please note that the folding was one of the main steps in Box-Least-Square algorithm.Their results showed that deep learning can play an important role for the detection of exoplanets. However, in these previous work, only one-dimension convolutional neural network (1D-CNN) was used. The input of an 1D-CNN model is a one-dimension array which has the flux values at different time of a light curve. When folding is used, one needs to add up all folded light curves and take averages for flux values. The signals of transits can be enhanced only when the folding period is exactly the same as the transit period. When one searches for new transits with unknown transit periods, the resolution of trial periods which are employed as folding periods needs to be very high. To solve this problem, the method of two-dimension convolutional neural network (2D-CNN), which was mainly used for pattern recognition, is explored here. All flux values of folded light curves can be kept and the transit signals would not be averaged out even when the folding period is different from the transit period. In this paper, both 1D-CNN and 2D-CNN deep learning models with phase folding will be constructed. Based on convolutional neural network (CNN), we study several deep learning models and compare their performances.In Section 2, the basic concept of machine learning, deep learning, and CNN will be introduced. In Section 3, five deep learning models are constructed based on convolutional neural network (CNN). The samples of light curves used as training, validation, and testing will be described in Section 4. The results and the demonstration will be presented in Section 5 and Section 6. Conclusions would be made in Section 7.
Artificial IntelligenceArtificial Intelligence (AI) is a computer program that can do some tasks automatically. For example, when we try to investigate whether there are planets moving around stars or not, we can give observed light-curve data of stars to an AI program. After the AI does some analysis, it will give an answer that there are planets or not around a particular star. Depending on the design of this AI program, in addition to showing the existence of planets, more parameters such as orbital p...