Table of contents

• Introduction
• Background Knowledge
• Environment
• File Guide
• Reference

Introduction

This project used a 1D CNN model to compare two different kinds of noise generated light curves.
The methods of generated light curves are following:

1. original noise data generated from dataset light curves in Kepler mission
2. noise data generated from quasi-period system by Pearson et al. (2019)

After constructing several training datasets by using those two methods with theoretic light curve formula in Mandel & Agol (2002), we trained our CNN model with K-fold Cross-Validation. Then we selected the best method to search possible transit light curves for Kepler dataset.

Background Knowledge

A planet that orbits a star outside the solar system is called an exoplanet.
There are many of methods to search exoplanets. The common methods are including: Doppler effect method, transit method, astrometry method and so on.
In this project, we used transit method and 1D CNN model to search whether there exists exoplanet candidates in Kepler Q1 dataset.

• Transit method:
  Because stars can emit stable light, planets cannot. Due to this property, when a plenet passes between a star and its observer, the bright emitted by this star drops. This is called a transit event.
  Transit event video source from NASA

Environment

This project is created with:

• Python version: 3.8
• Tensorflow-gpu version: 2.5.0
• Numpy version: 1.20.3
• Matplotlib version: 3.4.2
• astropy version: 4.3.1

To install quickly:

pip install -r requirements.txt

Notice:
If you want to use my code, please ensure package version especially tensorflow version. Sometimes when you used different version, it could be incompatible or some module have be deleted.

File Guide

• To preprocess kepler datas.
  Kplr/compare_KM.py - to record and compare magnitude of datas we had chosen from Kepler mission.
  Kplr/dev_analysis.py - This file is to analyze sigma of our kepler datas.We divided them into 4 group and hoped to have at least 50 in each group.
  Kplr/look_star.py - pre-processing datas from selected kepler mission.
  Kplr/read_fits.py - To convert .fits to .txt
• To generate our training datasets.
  generate_data_k.py - To generate training dataset using selected Kepler datas.
  generate_data_p.py - To generate training dataset using quasi-period system.
• CNN
  model_structure.py - Our 1D CNN model and training method(K-fold, early stop training)
  train.py - To train our model for each sample size of dataset and record training time.
• important and useful tools.
  mangol.py - the model to simulate transit event.
  tool.py - some common tools we often used.

Reference

• Thesis
  [1] 郭芷綺(2020)。以機器學習法搜尋系外行星的研究。國立清華大學碩士論文。
  [2] Mandel, K., & Agol, E. (2002). Analytic light curves for planetary transit searches. The Astrophysical Journal, 580:L171–L175, 2002 December 1
  [3] 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.
  [4] Yeh, Li-Chin & Jiang, Ing-Guey (2021). Searching for Possible Exoplanet Transits from BRITE Data through a Machine Learning Technique. Publications of the Astronomical Society of the Pacific, 133:014401 (12pp), 2021 January.
• Repo
  Pearson KA, Palafox L., Griffith CA, 2018, MNRAS, 474, 478