客觀的擊球技術動作類型辨識，在桌球領域幾乎是所有業餘愛好者們共通的需求，但是這項任務往往需要經驗豐富的教練來完成。為了解決此需求，眾多結合影像、感測器的研究與產品陸續被推出，但皆仍因為桌球領域的物理性限制，在成效上有所受限。

在此研究中，我們分析桌球拍擊球產生之聲音訊號，證實透過音訊處理技術，可以從擊球音訊中，推測得知對應之擊球技術類型；也利用聲音在時域上具較高解析度的特性，得以提取到使用攝影機與穿戴式感測器無法獲得之重要特徵，進行較精細的擊球技術類型區分。而在實務應用上，因為擊球的音訊前後必定常常伴隨著球與球桌接觸碰撞的聲音，為了避免這些碰撞聲干擾擊球音訊的分析，所以本研究中亦提出了一套區分擊球音訊與觸桌音訊的運算方式，藉此區分桌球與球桌碰撞聲。

本研究中，我們邀請台灣桌球成人國手選手，輔助建立桌球聲訊資料庫。目前資料庫內之音訊，包含擊球技術之音訊，與擊球觸桌之音訊資料，共1246筆。據本論文所提出，結合物理特性與機器學習之音訊處理演算法，在上述資料庫中，區分擊球音訊與觸桌音訊之演算法，準確率可達98.86%；區分擊球技術為扣殺、攻球、前衝、弧圈與切球五種技術動作，準確率可達90.25%。經量測驗證，此音訊處理演算法，可於1秒內完成高達每秒50個音訊事件以上之聲音訊號分析，說明此演算法具有可Real-time執行之可行性。

此音訊處理演算法系統在未來應用上，除了單獨使用之外，亦可與影像辨識系統或是慣性感測系統整合，架構更完善的系統，於情資蒐集、教學應用、競賽報導等情境，可提供更有效的工具。

Each athlete is eager to understand the efficiency of his skill to achieve higher performance. Conventionally it is achieved by the comments from coaches. Even though, without quantitative values limits the players' perception.

This research focuses on the recognition of table tennis skill using acoustic-characteristics when the ball is hit. Compared with the techniques like video analyses and wearable electronics that are proposed to understand the posture, using acoustics has several advantages. Most importantly, the higher sampling rate and resolution are helpful for understanding the moment of strokes. With such advantage, more detailed table tennis skills can be recognized.

In this research, by combining the physical-based knowledge and the machine learning techniques, we can distinguish the difference between ball-table and ball-paddle impacts, and five skills of Smash, Attack, Speedo, Loop, and Chop. The accuracy reaches 98.86% and 90.25% respectively. The processing time of the proposed system is less than 650ms for 50 events in 1 second, which makes real-time response possible.

Abstract............ ii
目錄................ iv
圖目錄.............. viii
表目錄.............. x

第一章 緒論..................................................2
1.1 研究動機.............................................2
1.2 困難與挑戰...........................................3
1.3 內容大綱.............................................3

第二章 文獻回顧與討論.........................................4
2.1 桌球技術自動化分類系統................................4
2.2 桌球相關之振動聲學研究................................6
2.3 章節總結.............................................8

第三章 實驗設計與桌球音訊資料庫建立...........................10
3.1 資料標籤類別定義....................................10
3.2 實驗器材與人力......................................11
3.2.1 錄音設備.......................................11
3.2.2 其他器材.......................................12
3.2.3 實驗人力.......................................13
3.3 實驗設計與資料蒐集流程..............................13
3.3.1 子資料集一 : 擊球聲與觸桌聲之資料蒐集............13
3.3.2 子資料集二 : 五種不同擊球技術之資料蒐集..........15
3.4 資料庫之內容組成...................................17
3.4.1 子資料集一 : 擊球聲與觸桌聲之資料組成............17
3.4.2 子資料集二 : 五種不同擊球技術之資料組成..........17
3.5 章節總結..........................................18

第四章 音訊處理演算法設計...................................20
4.1 音訊處理流程設計大綱................................20
4.2 事件偵測子系統之設計................................21
4.2.1 預濾波.........................................21
4.2.2 生成候選事件...................................21
4.2.3 事件篩選.......................................23
4.3 特徵提取子系統之設計................................24
4.3.1 濾波器設計.....................................24
4.3.2 頻域特徵提取設計................................25
4.3.3 時域特徵提取設計................................29
4.4 類別分類子系統之設計................................31
4.4.1 階段一分類器設計................................32
4.4.2 階段二分類器設計................................33
4.5 章節總結...........................................36

第五章 實驗結果與分析.......................................38
5.1 擊球與觸桌之音訊辨識成果............................38
5.1.1 擊球與觸桌分類之實驗結果.........................38
5.1.2 結果討論與分析..................................40
5.2 五種擊球技術之音訊辨識成果..........................44
5.2.1 One-Material-Out之交叉驗證結果與分析............44
5.2.2 10-Fold之交叉驗證結果與分析.....................47
5.2.3 交叉驗證準確率變因分析..........................51
5.2.4 誤判案例之成因分析..............................53
5.2.5 模型設計之優缺點總結............................53
5.3 系統Real-Time可行性評估...........................54
5.4 章節總結..........................................55

第六章 結論與未來發展.......................................56
6.1 結論..............................................56
6.2 未來應用...........................................56
6.3 未來工作...........................................57

參考文獻....................................................60

[1] C. Jordan, P. Renaud and M. Laurent, “Optical Flow Singularities for Sports Video Annotation: Detection of Strokes in Table Tennis,” MediaEval 2019 Workshop, Sophia Antipolis, France, 2019.
[2] P. E. Martin, B. P. Jenny, R. Péteri and J. Morlier, “Fine grained sport action recognition with Twin spatio-temporal convolutional neural networks: Application to table tennis,” Multimedia Tools and Applications, Apr, 2020.
[3] P. E. Martin, B. P. Jenny, R. Péteri and J. Morlier, “Sport Action Recognition with Siamese Spatio-Temporal CNNs: Application to Table Tennis,” International Conference on Content-Based Multimedia Indexing, La Rochelle, France, Sep, 2018, pp.1-6
[4] R. Liu, Z. Wang, X. Shi, H. Zhao, S. Qiu, J. Li and N. Yang, “Table Tennis Stroke Recognition Based on Body Sensor Network,” IDCS , 2019.
[5] S.M Lim, H.C Oh, J. Kim, J. Lee and J. Y. Park, “LSTM-Guided Coaching Assistant for Table Tennis Practice,” Sensors, vol.18, no. 12, 2018.
[6] B. Peter, H. Julian, S. Dominik and B. M. Eskofier, “Sensor-based stroke detection and stroke type classification in table tennis,” ISWC '15, New York, NY, USA, 2015, pp. 93–100.
[7] M. Sanusi, A. Khaleel, D. D. Mitri, B. Limbu and R. Klemke, “Table Tennis Tutor: Forehand Strokes Classification Based on Multimodal Data and Neural Networks,” Sensors, vol. 21, no. 9, 2021.
[8] 莊國良（2017）。桌球抽球時宜表現之探討:練習與比賽。國立臺灣師範大學體育學系博士論文，台北市。
[9] 鈴木一(1998)。桌球勝經。台北縣: 臺灣廣廈出版集團。
[10] 魏君達（2008）。桌球正手弧圈球技術動作分析。國立臺灣體育大學(桃園)運動技術研究所碩士論文，桃園縣。
[11] 郭嘉民（2006）。優秀桌球選手正手拉球之運動學分析。國立臺南大學運動與健康研究所碩士論文，台南市。
[12] Y. Kawazoe and D, Suzuki, “Prediction of table tennis racket restitution performance based on the impact analysis,” Theoretical and Applied Mechanics, vol. 52, pp. 163–174, 2003.
[13] Y. Kawazoe and D. Suzuki, “Impact prediction between a ball and racket in table tennis,” In A. Lee, J.-F. Kahn, and I. W. Maynard (Eds.), Science and racket sports III , pp. 134–139, 2004.q
[14] 麥亭萱（2020）。不同組合的桌球拍對球反彈表現之影響。臺北市立大學運動器材科技研究所碩士論文，臺北市
[15] 彭茜瑜（2006）。不同材質的桌球拍底板對打擊特性之探討。臺北巿立體育學院運動器材研究所碩士論文，臺北市。
[16] D. A. Russell, “Acoustics of ping-pong: Vibroacoustic analysis of table tennis rackets and balls,” Journal of Sports Sciences, vol. 36, no. 23, pp. 2644-2652, 2018.
[17] H. Lamb, "On the vibrations of a spherical shell," Proceedings London Mathematical Social, vol.1-13, no. 1, pp. 189-212, 1882
[18] W. E. Baker, "Axisymmetric modes of vibration of thin spherical shell, " Journal of the Acoustical Society of America, vol. 33, no. 12, 1961
[19] B. Zhang, W. Dou and L, Chen, "Ball Hit Detection in Table Tennis Games Based on Audio Analysis," 18th International Conference on Pattern Recognition (ICPR'06), pp. 220-223, 2006.
[20] 葉仁超（2006）。運動影片中的音訊事件偵測。國立清華大學資訊系統與應用研究所碩士論文，新竹市。
[21] D. Murray, (2002). Rebound of hollow elastic spheres. [Online]. Available:
http://www.geocities.ws/cofrest/md3.htm
[22] R. Cross, "Impact behavior of hollow balls," American Journal of Physics, vol. 82, no. 3, pp. 189–195, 2014