臺灣博碩士論文加值系統

本篇論文針對當使用心電圖訊號進行五種情緒分類時，如何藉由不同分類策略來提升情緒辨識的效能。欲辨識的情緒有五種：平常狀態(一般不受刺激的情緒狀態)、快樂、壓力、悲傷、生氣。受測者為十位男性與十位女性，使用的情緒刺激源為觀賞一段二至四分鐘的影片。
在使用二元分類器(binary classifier)進行多種類別分類時，需要結合多個二元分類器來進行分類，不同的結合方式各有優缺點，至今仍是一個還在研究的課題，一般在特徵挑選時，挑選出一組正確率最高的特徵進行分類，而本研究預期針對二元分類器組中，每一個分類器分別進行特徵挑選，選出多組特徵用在各個分類器，可以提升辨識正確率。研究中我們會對三種組合架構、兩種特徵選取方式、兩種分類器，搭配單組與多組特徵的方式進行實驗比較，使用到的組合架構有：OVO(one-against-one)、DAG(directed acyclic graph)以及本研究針對情緒辨識所提出的樹狀架構，共三種架構，在特徵挑選使用GA(genetic algorithm)與SFS(sequential forward selection)兩種，分類器特別使用一個二元分類器與一個非二元分類器來驗證上述方法，兩種分類器分別為SVM(support vector machine)與k-NN(k-nearest neighborhood)。
在使用三種分類架構，使用單及多組特徵，搭配兩種特徵選取法及使用兩種分類器的各種組合中，使用樹狀架構可以減少SVM分類時所使用之分類器個數亦維持相當的正確率，而使用多組特徵在上述任何的組合之中皆可增加辨識率，其中正確率最高的組合是OVO架構，搭配採用GA挑選多組特徵，並使用k-NN分類器，可以達到94%的正確率。

In this thesis, we study different classification strategies for emotion recognition based on electrocardiogram (ECG) to recognize five kinds of emotions, including neutral (non-stimulated state), happy, stress, sad, and anger. The participants consisted of 10 male and 10 female students who watched video programs of two to four minutes in length to stimulate distinct emotions.
Multiclass classification using binary classifiers need to combine several of binary classifiers. How to extend it effectively is still an ongoing research topic. In regular settings, one feature set is selected for all the binary classifiers to. In this study, we hypothesize that using unique feature set for the classifier can raise the accuracy for multiclass classification. Experiment were designed to the performance of combinations of three combining architectures, two feature selectors, and two kinds of classifications. The three architecture includes one-against-one, directed acyclic graph, and tree, which we propose for emotion recognition in this research. The two feature selectors are sequential forward selection and genetic algorithms. The two kinds of classifiers includes the support vector machine which is binary and the k-nearest neighborhood which is not binary classifier.
　　Comparing the performance of the above-mentioned different combinations, we found the combination of the tree architecture and the SVM classifier could reduce the number of classifiers yet retain acceptable accuracy. Combination of GA selector and OVO architecture using k-NN classifier with multiple feature sets can achieve the highest accuracy of 94%.

目錄
摘要 I
Abstract II
目錄 III
圖目錄 VI
表目錄 VIII
第一章 緒論 1
1.1 研究動機 1
1.2 相關文獻回顧 1
1.3 研究目標 2
1.4 論文架構 2
第二章 情緒辨識系統 4
2.1 情緒介紹 4
2.1.1 情緒的定義 4
2.1.2 情緒的分類 5
2.2 情緒辨識系統簡介 6
2.3 情緒辨識系統的類別 7
2.3.1 使用者相依的情緒辨識系統 7
2.3.2 使用者獨立的情緒辨識系統 8
2.4 生理訊號情緒辨識系統 8
2.4.1 生理訊號情緒辨識系統簡介 8
2.4.2 心電圖介紹 8
第三章 研究方法 12
3.1 生理訊號擷取 12
3.1.1 量測生理訊號的方法 12
3.1.2 訊號擷取的裝置 13
3.1.3 實驗流程 14
3.2 ECG特徵擷取 16
3.2.1時域特徵 17
3.2.2 HRV序列特徵 18
3.2.3 Poincare plot特徵 18
3.2.4基線區間特徵 21
3.2.5非線性特徵 21
3.2.6頻域特徵 23
3.2.7波形特徵 24
3.3特徵正規化 27
3.4 特徵選取 27
3.4.1逐次正向式搜尋法 27
3.4.2基因演算法 28
3.5 分類器 30
3.5.1 k-最近鄰居分類法 30
3.5.2支持向量機 31
3.6多分類架構 33
3.6.1一對其他 33
3.6.2一對一 34
3.6.3有向無環圖 35
3.6.4樹狀 35
3.6.5不平衡資料比例調整 39
3.6.6留一交叉驗證 40
第四章 實驗結果與討論 41
4.1各架構使用單組特徵搭配兩種分類器與選取方式之討論 41
4.1.1各架構使用單組特徵搭配SVM與兩種特徵挑選結果 41
4.1.2各架構使用單組特徵搭配k-NN與兩種特徵挑選結果 42
4.2三種架構使用多組特徵搭配兩種分類器與選取方式之討論 43
4.2.1各種分類架構使用多組特徵搭配SVM分類器與兩種特徵挑選 43
4.2.2各種分類架構使用多組特徵搭配k-NN分類器與兩種特徵挑選 44
4.3 一組與多組特徵影響之討論 45
4.3.1一與多組特徵對OVO與DAG兩種架構正確率與特徵數量影響 45
4.3.2一組與多組特徵對Tree架構之正確率與特徵數量影響 47
4.4個別情緒正確率之討論 47
4.5 樹狀架構超取樣有無之討論 48
4.6 樹狀重複類別之討論 48
4.7無特徵挑選不同架構正確率比較 49
4.8 經選取所得到特徵之討論 50
4.9 相關文獻比較 54
第五章 結論與未來發展 56
5.1 結論 56
5.2 未來發展 56
參考文獻 58
附錄一　特徵編號對照表 62
附錄二　特徵選取到的特徵 64

[1] “台灣設計波酷網”
http://www.boco.com.tw/ColumnistDetail.aspx?bid=B20090512000001
[2] Rosalind W. Picard, Affective Computing, The MIT Press, 1997
[3] K.H. Kim, S.W. Bang, and S.R. Kim, “Emotion Recognition System Using
Short-term Monitoring of Physiological Signals,” Medical & Biological Engineering & Computing, vol. 42, pp. 419-427, May 2004.
[4] Bin Wang, and G.Y. Liu, “An Experimental Study on Electrocardiography toward Emotion Recognition,” 2009 Sixth International Conference on Fuzzy Systems and Knowledge Discovery, vol. 1, pp. 104-108, Aug 2009.
[5] C.K. Lee, S.K. Yoo, Y.J. Park, N.H. Kim, K.S. Jeong, and B.C. Lee, “Using neural
network to recognize human emotions from heart rate variability and skin resistance,” Engineering in Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th Annual International Conference of the. IEEE, 2006.
[6] S. Cheong, S. H. Oh, and S.Y. Lee, “Support vector machines with binary tree architecture for multi-class classification,” Neural Info. Process.—Lett. Rev., vol. 2, no. 3, Mar. 2004.
[7] Fei Ben, and Jinbai Liu, “Binary tree of SVM: a new fast multiclass training and classification algorithm,” IEEE Trans Neural Netw., vol. 17, pp. 696-704, May 2006.
[8] K. Varpa, H. Joutsijoki, K. Lltanen, M. Juhola “Applying One-vs-One and One-vs-All Classifiers in k-Nearest Neighbour Method and Support Vector Machines to an Otoneurological Multi-Class Problem,” 23rd, European Federation of Medical Informatics; User centred networked health care; proceedings of MIE 2011. pp. 579-583, 2011.
[9]Chapelle Olivier, and S. Sathiya Keerthi, “Multi-class feature selection with support vector machines,” Proceedings of the American statistical association. 2008.
[10] Mehrabian, A., Russell, J. A., “An Approach to Environmental Psychology”, MIT Press, Cambridge, MA, 1974.
[11] J. Schwartz, P.R. Shaver, “Emotions and emotion knowledge in interpersonal
relations. In W. Jones & D. Perlman (Eds.), Advances in personal relationships”,
Greenwich, CT: JAI Press, Vol. 1, pp. 197-241, 1987.
[12] Ekman, P. and Friesen, W. V., “Constants across cultures in the face and
　emotion,” Journal of Personality and Social Psychology, vol. 17, pp
　124-129, 1971.
[13] C. E. Osgood, J. G. Suci, P.H. Tannenbaum , “The Measurement of Meaning,”
The University of Illinois Press, Urbana, 1957.
[14] Russell, J.A., Pratt, G., “A Description of The Affective Quality Attributed to
Environments”, Journal of Personality and Psychology, vol. 38, no. 2, pp. 311-322, 1980.
[15] C. Lisetti, F. Nasozin, “Affective intelligent car interfaces with emotion
recognition,” In Proceedings of 11th International Conference on Human Computer Interaction, vol. 2, pp. 1-10, July 2005.
[16] Busso Carlos, S.B. Lee, and Shrikanth Narayanan, “Analysis of emotionally salient aspects of fundamental frequency for emotion detection,” IEEE Trans. Audio, Speech, Language Process. vol. 17, pp. 582-596, Mar 2009.
[17] P. Ekman, and W. V. Friesen, “Constants across cultures in the face and
emotion,” Journal of Personality and Social Psychology, vol. 17, pp.
124-129, 1971
[18] E. Velten, “ A laboratory task for induction of mood states,” Behavior, 1968.
[19] Eun-Hye Jang, Byoung-Jun Park, Sang-Hyeob Kim, “Emotion classification
based on physiological signals induced by negative emotions: Discriminantion of
negative emotions by machine learning algorithm,” 2012 9th IEEE International Conference on Networking, Sensing and Control, pp.283-288, April 2012.
[20]周嘉陞, “心電圖學必備”, 合記圖書出版社, 2001.
[21]Elsevier Ltd. Boron & Boulpaep: Medical Physiology, Updated Edition.
www.studentconsult.com
[22] M. Brennan, M. Palaniswami, and Peter Kamen, “Do existing measures of Poincare plot geometry reflect nonlinear features of heart rate variability?,” IEEE Transactions on Biomedical Engineering, vol. 48, no. 11, pp. 1342-1347, Nov 2001.
[23] P. J. Lang, “Behavioral Treatment and Bio-Behavioral Assessment: Computer Appliances,” In J. H. Johnson J. B. Sidowski, & T. A. Williams (Eds.), Technology in Mental Health Care Delivery Systems, pp. 119-137, 1980.
[24] W. P. Rosalind, “Toward Machine Emotional Intelligence: Analysis of Affective Physiological State,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 23, no. 10, pp. 1175-1191, Oct. 2001.
[25] S. M. Pincus , “Heart rate control in normal and aborted-SIDS infants,” American Journal of Physiology, Vol. 33, pp. 638-646, 1991
[26]M.J. Katz, “Fractals and Analysis of Waveforms,” computers in biology and medicine 18, vol. 18, pp.145–156, 1988.
[27]T. Balli, and R. Palaniappan, “A Combined Linear & Nonlinear Approach for Classification of Epileptic EEG Signals,” IEEE Engineering in Medicine and Biology Conference on Neural Engineering, pp.714–717, May. 2009.
[28]I. Yalcin, and K. Mehmet, “Combining classical HRV indices with wavelet entropy measures improves to performance in diagnosing congestive heart failure,” Computers in Biology and Medicine 37, pp. 1502-1510, 2009.

[29]John Holland, “Adaptation in Natural an Artificial System,” University of Michigan Press, 1975.
[30] J.C. Platt, N. Cristianini and J. Shawe-Taylor, “Large Margin DAG’s for
Multiclass Classification, “Advances in Neural Information Processing
Systems”, Cambridge, MA: MIT Press., vol. 12, pp.547–553, 2000.
[31] S. Haykin, Neural networks: A Comprehensive Foundation, Prentice Hall, 2nd Ed., 1999.
[32] LibSVM: A Library for Support Vector Machine, C.-C. Chang and C. J.
Lin. (2001). http://www.csie.ntu.edu.tw/cjlin/libsvm [Online]
[33] N. V. Chawla, K. W. Bowyer, L. O. Hall, W. P. Kegelmeyer, “SMOTE: synthetic minority over-sampling technique,” Journal Of Artificial Intelligence Research, vol. 46, pp.321-357, Jun. 2001.
[34] E. M. Reza, Arash Rezaei, and M. B. Behrouz, “Comparison of Classification Methods Based on the Type of Attributes and Sample Size,” Journal of Convergence Information Technology, vol. 14, pp. 95-102, 2009.