在天文學中，我們常以光變曲線來找尋小行星的自旋週期。光變曲線是天體的亮度對應於時間的變化的一個函數。傳統的方法是對光變曲線作週期分析，並將光變曲線對可能的週期進行疊合後，再以天文學家自身專業與經驗來判斷該光變曲線圖是何種週期類別。但是近年來由於天文觀測技術的提升，獲取的天文資料數量也大幅增加，傳統人工判斷的方法曠日廢時不再可行，因此需要設計有效的方法來協助天文學家判斷該光變曲線的週期類別。
本論文將以泛星計畫 (Pan-STARRS, Panoramic Survey Telescope and Rapid Response System) 所觀測出的光變曲線作為實驗的資料來源，以二階的傅立葉函數擬合泛星計畫的小行星光變曲線，並得到一個 reduce 2對應頻率的頻譜圖，在該頻譜圖中可以發現一至數個頻率有極小值的reduce 2，將該光變曲線對這幾個頻率進行疊合而得一個疊合光變曲線，並以此疊合光變曲線作為週期類別之判斷。
由於小行星的形狀不是正圓球體，在疊合光變曲線中若呈現雙峰狀並且與基線擬合程度夠高，便可以認定找到該小行星之週期，若呈現單峰狀則認定找到小行星之半週期，其餘的情況便認定該光變曲線無法找到小行星週期。在此狀況下，雙峰狀的疊合光變曲線類似W，而單峰狀的則類似V或倒V，因此本論文導入監督式學習(supervised learning)技術並使用卷積類神經網路(CNN, convolutional neural network)設計一個類似人工判斷週期類別之工具，以疊合光變曲線為輸入，最後輸出該疊合光變曲線之類別，即W、V、或無類別。實驗結果顯示，此方式的準確度高且判別速度遠快於傳統的人工瀏覽。

In astronomical researches, the rotation periods of asteroids can be derived from their light curves which are the brightness as a function of time. Traditionally, a periodical analysis is performed on the light curve, and then astronomers determine the category of a possible period according to the folded light curve (i.e., a light curve folded to a particular period). This process was relied on human inspection, but it becomes very formidable due to the advancement in the technology of astronomical observation in the last decade that increases the volume of astronomical data set dramatically. Therefore, manual inspection is no longer feasible, and it is necessary to adopt an automatic method to replace the aforementioned time-consuming human review process.
In this research, we use the asteroid light curves obtained from the Panoramic Survey Telescope and Rapid Response System (Pan-STARRS). The light curves are fitted using the second-order Fourier series to find the possible rotation periods from the periodogram (i.e., the reduced 2 as a function of period). Then, a folded light curve of a certain period is generated. When a folded light curve shows a clear trend with a double-peak feature (i.e., similar to W), it is identified as a full rotation period. If a folded light curve only shows a single peak, a half rotation period is suggested (i.e., similar to V). The other cases are seen as no period found. Therefore, we deployed a deep learning technology, using convolutional neural network (CNN) as a network architecture, to construct a model to classify the folded light curves to W, V, and other shapes. From the study, we found that our model can precisely and yet much more effectively recognize a result consistent with that of human inspection.

[1] N. Kaiser, H. Aussel, B. E. Burke, H. Boesgaard, K. Chambers, M. R. Chun et al., "Pan-STARRS: a large synoptic survey telescope array", Survey and Other Telescope Technologies and Discoveries, Vol 4836, pp.154-165, December 2002.
[2] Pan-STARRS official website. http://pswww.ifa.hawaii.edu/pswww/
[3] Space Telescope Science Institute to Host Data from World's Largest Digital Sky Survey, Dec 19, 2016, http://hubblesite.org/news_release/news/2016-41/year/2016
[4] The Pan-STARRS1 data archive home page. https://panstarrs.stsci.edu/
[5] Pan-STARRS Sky Survey, Jan 28, 2019, http://hubblesite.org/image/4316/news_release/2019-12
[6] Light Curves and What They Can Tell Us. https://imagine.gsfc.nasa.gov /science/toolbox/timing1.html
[7] Russell and H.N, ”On the light variations of asteroids and satellites”, Astrophysical Journal, Vol 24, pp.1-18, July 1906.
[8] Fujiwara A., Kawaguchi J, Yeomans DK, et al., ” The Rubble-Pile Asteroid Itokawa as Observed by Hayabusa” , Science , Vol 312, Issue 5778, pp.1330-1334, June 2006.
[9] Harris and A.~W, “The Rotation Rates of Very Small Asteroids: Evidence for 'Rubble Pile' Structure”, Lunar and Planetary Institute Science Conference, Vol 27, pp. 493, March 1996.
[10] Chan-Kao Chang, Hsing-Wen Line, Wing-Huen Ip, et al., ”Searching for Super-fast Rotators Using the Pan-STARRS 1”, The Astrophysical Journal Supplement, Vol 241, Number1, February 2019.
[11] Y. LeCun, B. Boser, J. S. Denker, et. al., “Backpropagation Applied to Handwritten Zip Code Recognition”, Neural computation, Vol 1, Issue:4, pp541-551, December 1989.
[12] Yann LeCun, Léeon Bottou, Yoshua Bengio, and Patrick Haffner, ” Gradient-Based Learning Applied to Document Recognition“, Proceedings of the IEEE, Vol 86, Issue: 11, pp2278 – 2324, November 1998.
[13] 陳柏樺,「CNN結合GA用於手寫字元之辨識」, 國立高雄應用科技大學電機工程系碩士論, 2011
[14] 李兆健,「卷積神經網路應用於中文字手寫風格辨識」, 國立成功大學工程科學系碩士論文, 2017。
[15] MNIST dataset official website. http://yann.lecun.com/exdb/mnist/
[16] TensorFlow official website. https://www.tensorflow.org/
[17] Keras official website. https://keras.io/
[18] Reduced chi-squared statistic. https://en.wikipedia.org/wiki/Reduced_chi-squared_statistic
[19] A.Krizhevsky, I.Sutskever, and G.E.Hinton, “ImageNet Classification with Deep Convolutional Neural Networks”, NIPS'12 Proceedings of the 25th International Conference on Neural Information Processing Systems, pp1097-1105, Lake Tahoe, Nevada , December 2012.
[20] K.Simonyan, A.Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition” Proc. of IEEE Int. Conf. on Computer Vision and Pattern Recognition(CVPR), 1409.1556, June 2015.
[21] C.Szegedy, W.Liu, Y.Jia, P.Sermanet, S.Reed, D.Anguelov, D.Erhan, V.Vanhoucke, and A.Rabinovich, “Going Deeper with Convolutions”, Proc.of IEEE Int. Conf.on Computer Vision and Pattern Recognition(CVPR), pp.1-9, Boston,MA, June 2015.
[22] K.he, X.Zhang, S.Ren, and J.Sun, “Deep Residual Learning for Image Recognition”, Proc. of IEEE Int. Conf. on Computer Vision and Pattern Recognition(CVPR), pp.770-778, Las Vegas, NV, June 2016.
[23] Deep Face Recognition with Keras, https://sefiks.com/2018/08/06/deep-face-recognition-with-keras/
[24] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, R. Salakhutdinov, “Dropout: A Simple Way to Prevent Neural Networks from Overfitting”, Journal of Machine Learning Research, pp.1929-1958, June 2014.
[25] A. Mikołajczyk and M. Grochowski, “Data augmentation for improving deep learning in image classification problem”, pp.117-122, 2018 International Interdisciplinary PhD Workshop (IIPhDW), Swinoujście, Poland, May 2018.