臺灣博碩士論文加值系統

本論文提出一個以肌肉機械訊號(Mechanomyography, MMG)為訊息資料，用兩個置於手腕的三軸加速規建構而成的多通道手指動作辨識系統。硬體設置上有著簡易與方便的優點，並能夠辨識多達九種目標動作。研究中選用三種不同的特徵類別，包括一般時域特徵、多通道相關性特徵、以及相空間特徵。其中後兩者是MMG領域較少被使用到的特徵，在本論文中將其提出，並證實其有效性。分類器方面，則嘗試了線性支持向量機(Support vector machine, SVM)與k-最鄰近結點分類法 (k-nearest neighbor algorithm, kNN)，並發現這兩種分類器分別適合用於不同的情況。此外，我們也嘗試三種特徵選取方法：費雪鑑別分析(Fisher discriminant analysis, FDA)、逐次前饋式搜索法(Sequential forward selection, SFS)與逐次反向式搜索法(Sequential backward selection, SBS)，以對特徵進行分析探討。
考量到實際應用時會遭遇的各種情形，本實驗在分類時從許多方向著手，測試此系統面臨不同狀況時辨識率受到影響的程度。結果顯示，本系統在單人使用時可達90.74%辨識率，也能容忍多達十人共用，辨識率略降至82.5%。此外，考慮到實際使用的狀況，使用者可能會不經意作出與所設定之目標動作不同的手指移動，因此我們將系統進行修正，加入一簡單卻有效的分類步驟，使系統可以自動排解此不經意的動作，增進系統的強健性。而辨識度仍然可維持在單人90.04%，多人81.15%的正確率。然而，在各種情況的測試下，我們也驗證出幾個系統的限制。第一，在新使用者第一次操作此系統時，儘管系統已被他人訓練過，仍是需要重新訓練系統的。第二，相同使用者在不同天、或是有將感測器拆下再裝上的情況下使用時，也須再次進行一段簡易短暫的訓練。

In this thesis, we presented a multi-channel finger motion recognition system. Two tri-axial accelerometers attached to the wrist were used to measure mechanomyography signal as input data. This instrument setting is not only simple but also convenient, and could detect up to nine motions. We used three kinds of categories of features, including time-domain features, multi-channel correlation features, and phase space features in the study. Among these features, the multi-channel and phase space features are seldom appear in MMG research literature. We employed them and confirmed their abilities in MMG motion recognition. At for the classifiers, we tried the support vector machine and k-nearest neighbor algorithm, and realized that the two classifiers are suitable to different situations. Furthermore, we offered three feature selection methods: fisher discriminant analysis, sequential forward selection, and sequential backward selection, to select the most effective feature subset for the system.
Considering the conditions that may be confronted in different implementations, we tested our system by different experimental setups, to verify the accuracies in different environments. The results show that this system could reach 90.74% accuracy by single-user. The accuracy dropped slightly to 82.5% if used by multiple-users. Furthermore, in the practical use of the system, the users sometimes incidentally move their fingers that may become non-target motions in the study. Therefore, we modified and improved our system to be able to automatically detect the non-target motions, to promote the robustness of the system. After this improvement, the accuracies could still be hold at 90.04% under single-user and 81.15% under multiple-users situations. However, we do discover some limitations of our system. First, new users have to train the system before further operations. Second, users should quickly retrain system every time they reset the sensors.

口試委員會審定書 #
誌謝 i
中文摘要 ii
ABSTRACT iii
目錄 iv
圖目錄 vii
表目錄 viii
第1章 緒論 1
1.1 研究動機 1
1.2 相關文獻回顧 2
1.3 論文架構 3
第2章 研究方法 4
2.1 訊號前處理 4
2.1.1 去趨勢 4
2.1.2 自動動作區間切割 5
2.2 特徵擷取 6
2.2.1 時域特徵 6
2.2.2 多通道特徵 7
2.2.3 相空間特徵 8
2.3 分類器 11
2.3.1 k-最鄰近結點分類法 (k-nearest neighbor algorithm, kNN) 11
2.3.2 支持向量機 (Support vector machine, SVM) 13
2.4 特徵選取 16
2.4.1 費雪鑑別分析 16
2.4.2 逐次前饋式/反向式搜索法 17
第3章 實驗流程與設置 19
3.1 實驗設計 19
3.1.1 儀器設置與動作説明 19
3.1.2 受測者資料與量測協定 21
3.1.3 系統流程 22
3.2 實驗流程 24
3.2.1 訊號擷取 24
3.2.2 訊號前處理 25
3.2.3 特徵擷取 25
3.2.4 正規化 26
3.2.5 分類器 27
3.3 分類方法 27
3.4 特徵選取 29
第4章 實驗結果與探討 30
4.1 同天/單人/留一交互驗證 30
4.2 同天/多人/資料一半訓練一半測試 31
4.3 同天/多人/九人訓練一人測試 32
4.4 不同天/單人/留一交互驗證 33
4.5 同天/考慮進非目標動作之9+1情形 34
4.6 特徵選取結果之比較探討 36
4.6.1 費雪鑑別分析之結果 36
4.6.2 逐次前饋式搜索法之結果 39
4.6.3 逐次反向式搜索法之結果 41
4.6.4 三種特徵選取方法之比較 43
4.7 相關文獻比較 45
第5章 結論與未來發展 49
5.1 結論 49
5.2 未來發展 50
REFERENCE 52

[1]M. Watakabe, K. Mita, K. Akataki, and Y. Itoh, “Mechanical behavior of condenser microphone in mechanomyography”, Medical and Biological Engineering and Computing, Vol. 39, No. 2, pp. 195-201, 2011.
[2]M. Watakabe, K. Mita, K. Akataki, and Y. Itoh, “Technical aspects of mechanomyography recording with piezoelectric contact sensor”, Medical and Biological Engineering and Computing, Vol. 36, No. 5, pp. 557-61, 1988.
[3]C. Orizio, R. V. Baratta, B. H. Zhou, M. Solomonow, and A. Veicsteinas, “Force and surface mechanomyogram relationship in cat gastrocnemius”, Journal of Electromyography and Kinesiology, Vol. 9, No. 2, pp. 131-140, 1999.
[4]D. T. Barry, “Vibrations and sounds from evoked muscle twitches”, Electromyogr Clin Neurophysiol, Vol. 32, No. 1-2, pp. 35-40, 1992.
[5]F. M. Grimaldi, “Physico-mathesis de lumine, coloribus, et iride, aliisque adnexis libri duo”, Bologna, pp. 383, 1665.
[6]W. H. Wollaston, “On the duration of muscle action”, Philos. Trans. R. Soc. Lond. B. Biol. Sci., pp. 1-5, 1810.
[7]P. Madeleine and L. Arendt-Nielsen, “Experimental muscle pain increases Mechanomyographic signal activity during sub-maximal isometric contractions”, Journal of Electromyogr Kinesiol, Vol. 15, No. 1, pp. 27-36, 2005.
[8]P. Madeleine, P. Vedsted, A. K. Blangsted, G. Sjøgaard, and K. Søgaard, “Effects of electromyographic and Mechanomyographic biofeedback on upper trapezius muscle activity during standardized computer work”, Ergonomics, Vol. 49, No. 10, pp. 921-33, 2006.
[9]N. Alves and T. Chau, “The design and testing of a novel mechanomyogram-driven switch controlled by small eyebrow movements”, Journal of NeuroEngineering and Rehabilitation, vol. 7, pp. 22-31, 2010.
[10]J. Silva, W. Heim, and T. Chau, “A self-contained, mechanomyography-driven externally powered prosthesis”, Arch Phys Med Rehabil, vol. 86, pp. 2066-2070, Oct. 2005.
[11]H. B. Xie, Y. P. Zheng, and J. Y. Guo, “Classification of the Mechanomyogram signal using a wavelet packet transform and singular value decomposition for multifunction prosthesis control”, Physiol. Meas., Vol. 30, pp. 441-457, 2009.
[12]N. Alves and T. Chau, “Uncovering patterns of forearm muscle activity using multi-channel mechanomyography”, Journal of Electromyogr Kinesiol, Vol. 20, No. 5, pp. 777-786, Oct. 2009.
[13]X. Chen, X. Zhang, Z. Y. Zhao, J. H. Yang, V. Lantz, and K. Q. Wang, “Hand gesture recognition research based on surface EMG sensors and 2D-accelerometers”, IEEE conf. Wearable Computers, pp. 11-14, Oct. 2007.
[14]N. Alves and T. Chau, “Automatic detection of muscle activity from mechanomyogram signals: a comparison of amplitude and wavelet-based methods”, Physiol Meas., Vol. 31, No. 4, pp. 461-76, 2010.
[15]O. Malgina, J. Milenkovic, E. Plesnik, M. Zajc, and J. F. Tasic, “ECG signal feature extraction and classification based on R peaks detection in the phase space”, IEEE Conference and Exhibition (GCC), pp. 381-384, Feb. 2011.
[16]S. N. Yu and K. C. Hsu, “Detection of seizures in EEG using subband nonlinear parameters and genetic algorithm”, Computers in Biology and Medicine, Vol. 40, pp. 823-830, pp. 1770-1782, 2004.
[17]S. N. Rasband, “Chaotic Dynamics of Nonlinear Systems”, John Wiley & Sons, New York, 1990.
[18]K. T. Wu, “Bubble Coalescence from Mutiple Entry Nozzles in Fluidized Beds”, PhD Thesis, University College London, 1997.
[19]張淑清, “混沌時間序列重構相空間參數選取研究”, 物理學報, Vol.59, No. 3, pp. 1576-82, March, 2010.
[20]C. J. C. Burges, “A Tutorial on Support Vector Machines for Pattern Recognition”, Data Mining and Knowledge Discovery, Vol. 2, pp. 121-167, 1998.
[21]R. O. Duda, P. E. Hart, and D. G. Stork, “Pattern classification”, Wiley, New York, 2000.
[22]C. W. Hsu and C. J. Lin, “A comparison of methods for multi-class support vector machines”, IEEE Trans. on Neural Networks, Vol. 13, pp. 415-425, 2002.
[23]C. C. Chang and C. J. Lin, “LIBSVM, A library for support vector machines”, http://www.csie.ntu.edu.tw/~cjlin/libsvm/
[24]A. Whitney, “A direct method of nonparametric measurement selection”, IEEE Trans. Comput., Vol. 20, pp.1100-1103, 1971.
[25]S. J. Reeves and Z. Zhe, “Sequential algorithms for observation selection”, IEEE Trans. on Signal Processing, Vol. 47, No. 1, pp. 123-132, 1999.
[26]J. Silva and T. Chau, “Coupled microphone-accelerometer sensor pair for dynamic noise reduction in MMG signal recording”, Electronic Letters, Vol. 39, pp. 1496-1498, 2003.