我們使用深度學習法來建立兩種卷積神經網路模型，此兩種模型分別為全域特徵模型以及混合特徵模型。我們使用SuperWASP資料庫中的時間對光通量資料作為本研究中的雜訊光曲線資料；使用雜訊資料套用Mandel&Agol(2002)的非線性臨邊昏暗光曲線公式，使其產生出擁有凌星訊號的凌星光曲線。之後建構數個學習輸入數目的訓練集，並於兩種卷積神經網路訓練過程中皆使用交叉驗證法來訓練，最後比較全域特徵模型以及混合特徵模型的深度學習結果。

We use deep learning methods to construct two convolutional neural network models. These two models are the Global View Model and the Local Global View Model. In this thesis, we utilize time relative to the light flux data from the SuperWASP database as the noisy light curves for our research. The noisy data is subjected to the Mandel&Agol(2002) mentioned non-linear limb-darkening formula to generate transit light curves with transit signals. Subsequently, we construct multiple training sets with varying input sizes and employ cross-validation during the training process of both convolutional neural network models. Finally, we compare the deep learning results of the Global View Model and the Local Global View Model.

[1]https://exoplanetarchive.ipac.caltech.edu/docs/SuperWASPBulkDownload.html.
[2] S. Koning. Comparing convolutional neural networks and recurrent neural networks for exoplanet detection. Unpublished manuscript,Tilburg University, Tilburg, The Netherlands, 2018.
[3] K. Mandel and E. Agol. Analytic light curves for planetary transit searches. The Astrophysical Journal, 580(2):L171, 2002.
[4] C. J. Shallue and A. Vanderburg. Identifying exoplanets with deep learning: A five-planetresonant chain around kepler-80 and an eighth planet around kepler-90. The Astronomical Journal, 155(2):94, 2018.
[5] L.-C. Yeh and G. Jiang. Searching for possible exoplanet transits from brite data through a machine learning technique. Publications of the Astronomical Society of the Pacific, 133(1019):014401, 2020.