臺灣博碩士論文加值系統

目前在天文領域中，偵測小行星主要還是使用地對空的天文觀測望遠鏡進行密集的拍攝，並對拍攝到的圖片經前處理後進行數學之分析，但對於觀測圖片上肉眼難以發現的小行星，日前雖有研究以合成追蹤法(Synthetic Tracking)能夠發現其位置與軌跡，但仍須耗費非常大量的計算時間。因此，本論文提出一個能夠改善計算效率的架構，以幫助天文學家偵測此類訊號微弱的小行星。

本論文以在適當位置嵌入訊號微弱的昏暗小行星之星空模擬圖作為實驗的資料來源，以合成追蹤法為基礎並利用分散式系統和圖形處理器來加快運算速度，用不同的速度向量對圖片進行位移與疊合以增加小行星訊號的強度，並應用DAOPHOT光度測定軟體工具中的FIND演算法進行小行星之偵測，獲得圖片上可能存在小行星的概略速度及位置，透過分群整理後，再以更細的速度區間來偵測獲得準確的速度及位置，讓天文學家得以藉此迅速推算小行星之軌跡。

實驗結果顯示，以本論文提出的方法所偵測到的小行星，在偵測準確率的部份，在10組隨機生成的資料集中找到了90%的小行星，且移動速度在每秒10-2像素的量級下相當精準；在偵測效率的部分，對圖片位移與疊合的操作在圖形處理器的幫助下，節省了88%的運算時間，而FIND演算法的運算在Apache Spark的幫助下，平均可節省81%的時間；在實際的流程裡，分散式架構和圖形處理器平均分別可以節省64%和37%的總體運算時間。總結來說，透過分散式的平行運算和圖形處理器，此方法在保有一定的偵測準確率的條件下，最多可以成功地節省77%的運算時間，能夠為天文學家在小行星偵測的工作上達到更佳的效率。

Studies of asteroids have become more and more important since they are believed to help astronomical scientists to know the early evolution of the solar system. While the number of asteroids detected has grown numerously in recent years due to the significant improvement of photometry techniques, the detection of faint asteroids are still hard to achieve. In 2014, a method called Synthetic Tracking was brought up, which greatly enhances the performance of faint asteroids detection. Since it requires large-scale of workloads and costs plenty of time, we propose a new computing architecture to increase the performance.

Combining Synthetic Tracking with FIND algorithm in DAOPHOT software tool, our architecture shift-and-add simulated astronomical images to enhance signals of faint asteroids and detect their accurate velocities and locations. The system is established on distributed platform with GPU equipped, which largely speed up the performance of image processing.

Result shows that the accuracy of velocity detected is very high with 10-2 pixel per frame; the error of location is 2 pixels in average. For performance, the shift-and-add process saves 88% of time with GPU; the FIND algorithm saves 81% of time with help of Apache Spark. The total performance is enhanced 77% in average.

摘要 I
ABSTRACT II
目錄 III
圖目錄 V
表目錄 VI
一、 緒論 1
1-1 研究背景及動機 1
1-2 研究目的 1
1-3 論文架構 2
二、 文獻探討 3
2-1 合成追蹤法 3
2-2 DAOPHOT-FIND演算法 4
2-3 APACHE HADOOP 6
2-4 APACHE SPARK 8
2-5 DBSCAN 9
三、 研究方法 11
3-1 系統架構 11
3-2 資料前處理 12
3-2-1 去除離群值 12
3-2-2 減去背景值 12
3-3 篩選近似速度 13
3-3-1 設定速度區間 13
3-3-2 位移與疊合 13
3-4 以FIND演算法偵測小行星 14
3-5 小行星分群 15
3-6 配對小行星速度 16
四、 實驗 17
4-1 實驗資料集 17
4-2 實驗設備及環境 18
4-3 實驗整體結果 18
4-3-1 偵測結果 18
4-3-2 實驗準確率與運算效率評估 26
五、 結論及未來展望 31
參考文獻 32

[1] NASA JPL Center for Near Earth Object Studies. Available: https://cneos.jpl.nasa.gov/
[2] Shao, M. et al., “Finding Very Small Near-Earth Asteroids Using Synthetic Tracking”, The Astrophysical Journal, 782:1(10pp), February 2014
[3] Zhai, C. et al., “Detection of A Faint Fast-Moving Near-Earth Asteroid Using the Synthetic Tracking Technique”, The Astrophysical Journal, 792:60(14pp), September 2014
[4] Harding, L. K. et al., “CHIMERA: a wide-field, multi-colour, high-speed photometer at the prime focus of the Hale telescope”, Monthly Notices of the Royal Astronomical Society, Vol. 457, Issue 3, pp. 3036-3049, April 2016
[5] Zhai, C. et al., “Synthetic Tracking Using ZTF Deep Drilling Data Sets”, The Astronomical Society of the Pacific, 132:064502 (8pp), June 2020
[6] Stetson, P. B., “DAOPHOT: A COMPUTER PROGRAM FOR CROWDED-FIELD”, The Astronomical Society of the Pacific, Vol. 99, 191-222, March 1987
[7] Pérez, R. H. et al. (2016) “Improving Performance of DAOPHOT-FIND by Using GPU Architecture.”, ISUM 2015: High Performance Computer Applications, 240-249, Mexico City, Mexico, March 2015
[8] Apache Hadoop Official Website. Available: https://hadoop.apache.org/.
[9] Apache Spark Official Website. Available: http://spark.apache.org/
[10] Ester, M. et al., “A density-based algorithm for discovering clusters in large spatial databases with noise”, Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96), AAAI Press. pp. 226–231., August 1996