在本論文中，我們先建造了一個基礎模型去檢測光曲線，並且一一去比較卷積神經網絡參數，得到一個最佳模型，然後再利用最佳之模型去比較干擾參數不同的情形下準確率的差別，可發現干擾參數較大準確率較差，因此我們討論干擾參數較大時如何去改善其準確率，並在最後一章做總結。

In this thesis, we first build a basic model to detect the light curve, and compare the Convolutional Neural Network parameters one by one to get an optimal model, and then use this model to compare the difference in accuracy under different parameters. It can be found that the poor accuracy is caused by the large noise, so we discuss how to improve the accuracy when the noise parameter is large. Finally, we make a summary in the last chapter.

致謝.....I
摘要.....II
Abstract.....III
第一章 簡介.....1
第二章 神經網絡與凌日模型介紹.....3
2.1類神經網絡(Neural Network).....3
2.1.1神經細胞.....3
2.1.2類神經網絡運作.....3
2.1.3損失函數(Cost/Loss Function).....5
2.1.4感知器演算法(Perception Learning Algorithm).....6
2.2卷積神經網絡.....8
2.2.1卷積運算(Convolution).....9
2.2.2池化(Pooling).....10
2.2.3展平(Flattening).....11
2.2.4丟棄法(Dropout).....12
2.2.5全連接層(Full connection).....12
2.3凌日模型(Transit model).....12
第三章 參數變化的比較.....20
3.1激活函數測試.....21
3.2 kernel比較.....22
3.3層數比較.....23
3.4學習優化器(Optimizer)比較.....24
3.5訓練樣本數比較.....25
第四章 模擬結果.....27
4.1模型A.....29
4.2模型B.....32
4.3模型C.....35
4.4模型比較.....37
4.5模型改善.....39
第五章 結論.....41
參考文獻.....42
附錄.....43
A、學習優化器[6].....43
A.1梯度下降法(Gradient Descent).....43
A.2 RMSProp 方法.....44
A.3 Adam 方法.....44
B、檢測有無行星通過之CNN模型程式碼.....46

[1] Chintarungruangchai, P., & Jiang, G. (2019). Detecting Exoplanet Transits through Machine-learning Techniques with Convolutional Neural Networks. Publications of the Astronomical Society of the Pacific, 131(1000), 064502.
[2] Mandel, K., & Agol, E. (2002). Analytic light curves for planetary transit searches. The Astrophysical Journal Letters, 580(2), L171.
[3] Nikhil Ketkar, Deep Learning with Python, Springer,2017
[4] Pearson, K. A., Palafox, L., & Griffith, C. A. (2018). Searching for exoplanets using artificial intelligence. Monthly Notices of the Royal Astronomical Society, 474(1), 478-491.
[5] https://mropengate.blogspot.com/2015/06/ch15-4-neural-network.html
[6] https://github.com/PetarV-/TikZ/tree/master/Dropout