脑电信号在情感识别中的研究
|
摘要
情感计算是实现高级人机交互的关键技术之一,是人工智能领域中日益受到关注的一个研究方向。情感识别是情感计算的一个重要组成部分,包括语音、面部表情、文本、姿势和生理信号识别等方面,其中生理信号方面的情感识别研究是最困难的。基于生理信号的情感识别,主要研究对象是不同情感状态下的脑电、肌电、心电、皮肤电导、皮肤电阻、皮温、光电脉搏、呼吸信号等。美国MIT媒体实验室Picard教授所带领的情感计算研究小组主要是分析肌电(EMG)、呼吸作用(RSP)、皮肤电反应(SC)和血容量搏动(BVP)四种生理信号,率先从这些生理信号中提取各种特征来进行情感状态识别的研究,并证明应用生理信号来进行识别情感状态的方法是可行的。
脑电信号是大脑内部亿万神经元活动在大脑皮层的综合反映,能够直接反映大脑的活动情况。思维状态的不同和各种情绪的变化会在不同的大脑皮层位置反映出不同的脑电信号,因此,脑电信号含有丰富的有用信息,如何有效的处理脑电信号和提取其中的有用信息,对于情感状态识别的研究具有重要意义。然而由于脑电信号是一种产生机理相当复杂的非平稳随机信号,其数据的采集过程相当复杂,而且容易受外界环境和心电、肌电等其他生理信号的干扰,国内外将脑电信号用于情感状态识别中的研究与其他生理信号(如心电、肌电、皮肤电等)相比非常少。从德国Augsburg大学计算科学研究所采集的生理信号数据中可以看出,他们并没有采集脑电信号的数据。
论文在前人研究基于其他生理信号的情感识别的基础上,对脑电信号进行了情感识别方面的研究,主要研究内容如下：
1)脑电信号的数据采集：我们实验室通过采用美国Biopac公司提供的多导生理记录仪MP150同时采集包括脑电图(EEG)、肌电图(EMG)、心电图(ECG)、血容量搏动(BVP)、皮肤电反应(GSR)和呼吸作用等六种生理信号数据。实验总共采集了244个被试的生理信号,所有被试均来自西南大学在校大一学生；
2)原始脑电信号的预处理：通过分析脑电信号四种基本节律的特点,论文将脑电信号的β波节律作为情感识别研究的对象,用小波包变换从脑电信号中提取了β波节律,并对其进行功率普分析,验证了将脑电信号的β波用于情感识别研究的可行性；
3)脑电信号的特征提取：将用小波包变换提取的脑电信号的β波及其所包含的小波包分解结点的小波包系数进行计算,计算出脑电信号的各种统计值作为原始特征。愤怒、厌恶、恐惧、悲伤、高兴和惊奇这6种情感类型对应6种类别,愤怒样本个数为228,厌恶样本个数为96,恐惧样本个数为230,悲伤样本个数为238,高兴样本个数为206,惊奇样本个数为191,总共1189个样本；
4)脑电信号的特征选择：采用禁忌搜索算法用于情感生理信号的特征选择,将fisher作为情感模式的分类器,利用分类器的准确率来评价特征组,保留有利于准确率提高的特征组合,剔除使得准确率下降的特征组合,以此来评价特征组合的优劣。采用这种方式对情感的脑电信号特征进行选择,以便更有效地识别情感；
5)论文最后通过在一对一和一对多的情感状态识别中进行实验仿真。
论文通过从脑电信号中提取β波节律,并从β波中提取出相关的特征,然后利用禁忌搜索算法进行特征选择,最后在一对一和一对多的情感状态识别中进行实验仿真。从实验仿真结果分析得知,实验得到了比较满意的情感识别效果以及对识别特定情感有较大贡献的一些特征集,证实了我们所提取的脑电信号的特征用于情感识别是有一定识别效果的,将脑电信号的β波节律特征用于情感识别是可行的。
Affective computing is one of the key technologies to achieve high-level huma-computer interaction, and a research direction which is the increasing attention to the field of artificial intelligence. Emotion recognition is an integral part of affective computing, including voice, facial expressions, text, gesture, and physiological signal recognition, etc, which the emotion recognition of physiological signals is most difficult.The main object of emotion recognition based on physiological signals is different from emotional state of the EEG, EMG, ECG, skin conductance, skin resistance, skin temperature, optical pulse and respiration signals.The United States MIT Media Lab Affective Computing Research Group led by Professor Picard is to analyze electromyography (EMG), respiration (RSP), GSR (SC) and blood volume pulse (BVP). They are the first to extract features from these physiological signals to research emotion recognition and testified that it is feasible to recognize emotion from physiological signals.
EEG is the hundreds of millions of neurons within the brain activity in a comprehensive reflection of the cerebral cortex can be a direct reflection of brain activity. The different state of mind and a variety of emotional changes in the cerebral cortex in different locations reflect different brain signals, therefore, the brain signals are rich in useful information, how to deal effectively with these brain signals and extract features from them is very important to the emotional stae recognition.However, because the brain produces electrical signals is a very complex mechanism of non-stationary random signal, its data collection process is quite complicated and vulnerable to the external environment and the ECG, EMG and other physiological signals, the EEG at home and abroad used to identify the research of emotional state recognition is very few compared to other physiological signals(such as ECG, EMG, SC,etc.).From Germany, Augsburg University Institute of Computing Scienct physiological signal data collected, we can see that they did not collect EEG data.
This paper studied the research of EEG signal in emotion recognition based on the previous research results which are based on other physiological signals in emotion recognition, the main research contents are as follows:
(1) Cllection of EEG data:our laboratory through the use of the MP150 multi-channel physiological recorder which is provided by the United States Biopac Company to collected six physiological signal data, including electroencephalography (EEG), electromyography (EMG), electrocardiogram (ECG), blood volume pulse (BVP), skin electrical response (GSR) and respiration. We collected physiological signals data from the total of 244 subjects, all the subjects were all from the Southwest University freshman in school students;
(2) Original EEG preprocessing:this paper will put theβ-wave rhythm of EEG as the object of emotion recognition by analyzing the characteristics of the four basic rhythms, use wavelet packet transform to extract signals from theβ-wave rhythm of brain, and analysis its power spectrum to testified that it is feasible to recognize emotion from physiological signals with theβ-wave rhythm of EEG;
(3) Extraction of EEG features:theβ-wave rhythm, which is extracted form EEG by the use of wavelet packet transform, and its wavelet packet nodes of wavelet packet decomposition coefficients are used to calculate the the various statistics of EEG characteristics as the original features.Anger, disgust, fear, sadness, joy and surprise are corresponding to six kinds of categories of emotion, the sample number of anger is 228, the sample number of disgust is 96, the sample number of fear is 230, the sample number of sad is 238, the sample number of joy 206, the sample number of surprise is 191, the total sample number is 1433;
(4) Selection of EEG features:the tabu search algorithm is used to select the features of affective physiological signals, the fisher classifier is used as an emotional pattern classifier and its classification accuracy is used to evaluate the feature combination, that helps to retain the feature combination which is helpful to increase the accuracy rate and excluding the feature combination which declines the accuracy rate, in order to assess the strengths and weaknesses of the feature combination. In this way, we select features from EEG signals in order to identify the emotions effectively;
(5) Finally, papers make the experimental simulation of emotional state recognition in the way of one-to-one and one-to-many.
Papers extracted P-wave rhythm from EEG signals, and extracted related features from the P-wave, and then using tabu search algorithm for feature selection, in the end made the the experimental simulation of emotional state recognition in the way of one-to-one and one-to-many. Analysis the simulation results from the experiment, we has been relatively satisfied with the emotional stae recognition effect and got some feature set which has a greater contribution to identify a specific emotion.lt testified that the EEG features we extracted for emotion recognition got a certain recognition effect and it is feasible to recognize emotion with EEG signals.
脑电信号是大脑内部亿万神经元活动在大脑皮层的综合反映,能够直接反映大脑的活动情况。思维状态的不同和各种情绪的变化会在不同的大脑皮层位置反映出不同的脑电信号,因此,脑电信号含有丰富的有用信息,如何有效的处理脑电信号和提取其中的有用信息,对于情感状态识别的研究具有重要意义。然而由于脑电信号是一种产生机理相当复杂的非平稳随机信号,其数据的采集过程相当复杂,而且容易受外界环境和心电、肌电等其他生理信号的干扰,国内外将脑电信号用于情感状态识别中的研究与其他生理信号(如心电、肌电、皮肤电等)相比非常少。从德国Augsburg大学计算科学研究所采集的生理信号数据中可以看出,他们并没有采集脑电信号的数据。
论文在前人研究基于其他生理信号的情感识别的基础上,对脑电信号进行了情感识别方面的研究,主要研究内容如下：
1)脑电信号的数据采集：我们实验室通过采用美国Biopac公司提供的多导生理记录仪MP150同时采集包括脑电图(EEG)、肌电图(EMG)、心电图(ECG)、血容量搏动(BVP)、皮肤电反应(GSR)和呼吸作用等六种生理信号数据。实验总共采集了244个被试的生理信号,所有被试均来自西南大学在校大一学生；
2)原始脑电信号的预处理：通过分析脑电信号四种基本节律的特点,论文将脑电信号的β波节律作为情感识别研究的对象,用小波包变换从脑电信号中提取了β波节律,并对其进行功率普分析,验证了将脑电信号的β波用于情感识别研究的可行性；
3)脑电信号的特征提取：将用小波包变换提取的脑电信号的β波及其所包含的小波包分解结点的小波包系数进行计算,计算出脑电信号的各种统计值作为原始特征。愤怒、厌恶、恐惧、悲伤、高兴和惊奇这6种情感类型对应6种类别,愤怒样本个数为228,厌恶样本个数为96,恐惧样本个数为230,悲伤样本个数为238,高兴样本个数为206,惊奇样本个数为191,总共1189个样本；
4)脑电信号的特征选择：采用禁忌搜索算法用于情感生理信号的特征选择,将fisher作为情感模式的分类器,利用分类器的准确率来评价特征组,保留有利于准确率提高的特征组合,剔除使得准确率下降的特征组合,以此来评价特征组合的优劣。采用这种方式对情感的脑电信号特征进行选择,以便更有效地识别情感；
5)论文最后通过在一对一和一对多的情感状态识别中进行实验仿真。
论文通过从脑电信号中提取β波节律,并从β波中提取出相关的特征,然后利用禁忌搜索算法进行特征选择,最后在一对一和一对多的情感状态识别中进行实验仿真。从实验仿真结果分析得知,实验得到了比较满意的情感识别效果以及对识别特定情感有较大贡献的一些特征集,证实了我们所提取的脑电信号的特征用于情感识别是有一定识别效果的,将脑电信号的β波节律特征用于情感识别是可行的。
Affective computing is one of the key technologies to achieve high-level huma-computer interaction, and a research direction which is the increasing attention to the field of artificial intelligence. Emotion recognition is an integral part of affective computing, including voice, facial expressions, text, gesture, and physiological signal recognition, etc, which the emotion recognition of physiological signals is most difficult.The main object of emotion recognition based on physiological signals is different from emotional state of the EEG, EMG, ECG, skin conductance, skin resistance, skin temperature, optical pulse and respiration signals.The United States MIT Media Lab Affective Computing Research Group led by Professor Picard is to analyze electromyography (EMG), respiration (RSP), GSR (SC) and blood volume pulse (BVP). They are the first to extract features from these physiological signals to research emotion recognition and testified that it is feasible to recognize emotion from physiological signals.
EEG is the hundreds of millions of neurons within the brain activity in a comprehensive reflection of the cerebral cortex can be a direct reflection of brain activity. The different state of mind and a variety of emotional changes in the cerebral cortex in different locations reflect different brain signals, therefore, the brain signals are rich in useful information, how to deal effectively with these brain signals and extract features from them is very important to the emotional stae recognition.However, because the brain produces electrical signals is a very complex mechanism of non-stationary random signal, its data collection process is quite complicated and vulnerable to the external environment and the ECG, EMG and other physiological signals, the EEG at home and abroad used to identify the research of emotional state recognition is very few compared to other physiological signals(such as ECG, EMG, SC,etc.).From Germany, Augsburg University Institute of Computing Scienct physiological signal data collected, we can see that they did not collect EEG data.
This paper studied the research of EEG signal in emotion recognition based on the previous research results which are based on other physiological signals in emotion recognition, the main research contents are as follows:
(1) Cllection of EEG data:our laboratory through the use of the MP150 multi-channel physiological recorder which is provided by the United States Biopac Company to collected six physiological signal data, including electroencephalography (EEG), electromyography (EMG), electrocardiogram (ECG), blood volume pulse (BVP), skin electrical response (GSR) and respiration. We collected physiological signals data from the total of 244 subjects, all the subjects were all from the Southwest University freshman in school students;
(2) Original EEG preprocessing:this paper will put theβ-wave rhythm of EEG as the object of emotion recognition by analyzing the characteristics of the four basic rhythms, use wavelet packet transform to extract signals from theβ-wave rhythm of brain, and analysis its power spectrum to testified that it is feasible to recognize emotion from physiological signals with theβ-wave rhythm of EEG;
(3) Extraction of EEG features:theβ-wave rhythm, which is extracted form EEG by the use of wavelet packet transform, and its wavelet packet nodes of wavelet packet decomposition coefficients are used to calculate the the various statistics of EEG characteristics as the original features.Anger, disgust, fear, sadness, joy and surprise are corresponding to six kinds of categories of emotion, the sample number of anger is 228, the sample number of disgust is 96, the sample number of fear is 230, the sample number of sad is 238, the sample number of joy 206, the sample number of surprise is 191, the total sample number is 1433;
(4) Selection of EEG features:the tabu search algorithm is used to select the features of affective physiological signals, the fisher classifier is used as an emotional pattern classifier and its classification accuracy is used to evaluate the feature combination, that helps to retain the feature combination which is helpful to increase the accuracy rate and excluding the feature combination which declines the accuracy rate, in order to assess the strengths and weaknesses of the feature combination. In this way, we select features from EEG signals in order to identify the emotions effectively;
(5) Finally, papers make the experimental simulation of emotional state recognition in the way of one-to-one and one-to-many.
Papers extracted P-wave rhythm from EEG signals, and extracted related features from the P-wave, and then using tabu search algorithm for feature selection, in the end made the the experimental simulation of emotional state recognition in the way of one-to-one and one-to-many. Analysis the simulation results from the experiment, we has been relatively satisfied with the emotional stae recognition effect and got some feature set which has a greater contribution to identify a specific emotion.lt testified that the EEG features we extracted for emotion recognition got a certain recognition effect and it is feasible to recognize emotion with EEG signals.
引文
[1]Jianhua Tao and Tieniu Tan. Affective Computing:A Review, First International Conference [A]. ACII2005 [C]. Beijing. China. Oct 2005,982～993.
[2]Ekman P. An argument for basic emotion [J]. Cognition and Emotion.1992,6, 169～200.
[3]Stemmler G., Heldmann M., Pauls C. A., et al. Constraints for emotion specificity in fear and anger:the context counts [J]. Psychophysiology.2001,38,275～291.
[4]Elias Vyzas and Rosalind W. Picard. Affective Pattern Classification. Perceptual Computing Section Technical Report, No.473. M.I.T. Media Laboratory.1998.
[5]Picard R. W., Healey J. Affective Wearable [A]. In Proceedings of the First International Symposium on Wearable Computers [C]. Cambridge, MA, Oct 1997, 1:231-240.
[6]Ark W., Dryer D. C. and Lu D. J. The emotion mouse [A].8th Int. Conf. Human-computer Interaction [C].1999,453-458.
[7]罗森林,潘丽敏.情感计算理论与技术[J].系统工程与电子技术.2003,25(7),905～908.
[8]张颖,罗森林.情感建模与情感识别.计算机工程与应用[J].2003,(33),98～102.
[9]Kim K H, Bang S W, Kim S R. Emotion recognition system using short-term monitoring of physiological singnals.[J] Med Biol Eng Comput,2004,42,419-427.
[10]Jocelyn Scheirer, Raul Fernandez, et al. Frustrating the User on Purpose:A Step Toward Building an Affective Computer. M.I.T. Media Laboratory Perceptual Computing Section Technical Report No.509.
[11]日本文部省重点研究领域：情感信息处理的信息学、心理学研究.成果报告书,1999.
[12]Lan Li, Ji-hua Chen. Emotion Recognition Using Physiological Signals. Advances in Artificial Reality and Tele-Existence.0302-9743 (Print) 1611-3349 (Online),Volume 4282/2006,P 437～446.
[13]陈继华,李岚,钱坤喜.基于多生理信号的情绪初步识别[J].生物医学工程研究2006年03期.
[14]徐景波,孟昭兰,王丽华.正负性情绪的自主生理反应实验研究[J].心理科学.1995,18(3),134～139.
[15]Haag A., Goronzy S., Schaich P., Williams J. Emotion Recognition Using Bio-Sensors:First Step Towards an Automatic System [J]. Affective Dialogue Systems. Kloster Irsee, Germany.2004,3068,36～48.
[16]Nasoz F., Alvarez K., Lisetti C. L., Finkelstein N. Emotion Recognition from Physiological Signals for Presence Technologies [J]. International Journal of Cognition, Technology and Work, Special Issue on Presence,2003,6(1),1-32.
[17]刘贤敏,刘唱.中国古典音乐诱发情绪的生理活动研究[D].南京师范大学.2006,5.
[18]Wagner J, Kim J, Andre E. From Physiological Signals to Emotions: Implementing and Comparing Selected Methods for Feature Extraction and Classification [A]. In IEEE International Conference on Multimedia & Expo (ICME 2005) [C].2005,940～943.
[19]Lan Li and Ji-hua Chen. Emotion Recognition Using Physiological Signals [A]. ICAT 2006 [C]. LNCS 4282,2006,437-446.
[20]Mi-hee Lee, Gyunghye Yang, Hyoung-Ki Lee and Seokwon Bang. Development Stress monitoring System based on Personal Digital Assistant (PDA) [A]. Proceedings of the 26th Annual International Conference of the IEEE EMBS San Francisco [C]. CA, USA,2004,2364-2367.
[21]Timo Partala, Veikko Surakka. Pupil size variation as an indication of affective processing[J]. International Journal of Human Computer Studies.2003, 59(1),185～198.
[22]Jing Zhai and Armando Barreto. Stress Detection in Computer Users Based on Digital Signal Processing of Noninvasive Physiological Variables [A]. Proceedings of the 28th IEEE EMBS Annual International Conference [C]. New York, USA.2006,1355～1358.
[23]Julio C. D. Conway, Claudionor J. N. Coelho Jr., et al. Wearable Computer as a Multi-parametric Monitor for Physiological Signals [J]. IEEE 2000,236～242.
[24]Healey J. A. Wearable and Automotive Systems for Affect Recognition from Physiology. [PhD thesis] MIT, Cambridge, MA,2000.
[25]韩伟,张泾周,刘园园,张光磊.脑电节律提取的小波变换方法[J].中国医疗器械杂志,2006,30(1),19～21.
[26]葛哲学沙威.小波分析理论与MATLAB R2007实现[M].北京：电子工业出版社,2007.10.
[27]C.Sidney Burrus等著.程正兴译.小波与小波变换导论／(美)[M].北京：机械工业出版社,2007.10.
[28]Glover F., Future paths for integer programming and links to artificial intelligence, Computers and Operations Research,1986,13,533～549.
[29]Glover F. and M.Tabu Search. Boston, Kluwer Academic Publishers,1997.
[30]邢文训,谢金星.现代优化计算方法[M].清华大学出版社.2005,51-68.
[31]王凌.智能优化算法及其应用[M].北京：清华大学出版社,施普林格出版社.2010.10.
[32]Randall S. Sexton, Bahrain Alidaee, Robert E. Dorsey, and John D.Johnson, Global Optimization for Artificial Neural Networks:A Tabu Search Application, European Journal of Operational Research,1998,106, pp,570～584.
[33]Hughes J R. EEG in clinical practice. Butterworth-Heinemann,1994.
[34]Shi Y., Eberhart R C.A modified swarm optimizer [A].IEEE International Conference of Evolutionary Computation [C]. Anchorage, Alaska:IEEE Press, 1998.
[35]边肇祺,张学工.模式识别[M].清华大学出版社.2000,176-210.
[36]Sergios Theodoridis and Konstantinos Koutroumbas. Pattern Recognition, Second Edition,2004,106～124.
[2]Ekman P. An argument for basic emotion [J]. Cognition and Emotion.1992,6, 169～200.
[3]Stemmler G., Heldmann M., Pauls C. A., et al. Constraints for emotion specificity in fear and anger:the context counts [J]. Psychophysiology.2001,38,275～291.
[4]Elias Vyzas and Rosalind W. Picard. Affective Pattern Classification. Perceptual Computing Section Technical Report, No.473. M.I.T. Media Laboratory.1998.
[5]Picard R. W., Healey J. Affective Wearable [A]. In Proceedings of the First International Symposium on Wearable Computers [C]. Cambridge, MA, Oct 1997, 1:231-240.
[6]Ark W., Dryer D. C. and Lu D. J. The emotion mouse [A].8th Int. Conf. Human-computer Interaction [C].1999,453-458.
[7]罗森林,潘丽敏.情感计算理论与技术[J].系统工程与电子技术.2003,25(7),905～908.
[8]张颖,罗森林.情感建模与情感识别.计算机工程与应用[J].2003,(33),98～102.
[9]Kim K H, Bang S W, Kim S R. Emotion recognition system using short-term monitoring of physiological singnals.[J] Med Biol Eng Comput,2004,42,419-427.
[10]Jocelyn Scheirer, Raul Fernandez, et al. Frustrating the User on Purpose:A Step Toward Building an Affective Computer. M.I.T. Media Laboratory Perceptual Computing Section Technical Report No.509.
[11]日本文部省重点研究领域：情感信息处理的信息学、心理学研究.成果报告书,1999.
[12]Lan Li, Ji-hua Chen. Emotion Recognition Using Physiological Signals. Advances in Artificial Reality and Tele-Existence.0302-9743 (Print) 1611-3349 (Online),Volume 4282/2006,P 437～446.
[13]陈继华,李岚,钱坤喜.基于多生理信号的情绪初步识别[J].生物医学工程研究2006年03期.
[14]徐景波,孟昭兰,王丽华.正负性情绪的自主生理反应实验研究[J].心理科学.1995,18(3),134～139.
[15]Haag A., Goronzy S., Schaich P., Williams J. Emotion Recognition Using Bio-Sensors:First Step Towards an Automatic System [J]. Affective Dialogue Systems. Kloster Irsee, Germany.2004,3068,36～48.
[16]Nasoz F., Alvarez K., Lisetti C. L., Finkelstein N. Emotion Recognition from Physiological Signals for Presence Technologies [J]. International Journal of Cognition, Technology and Work, Special Issue on Presence,2003,6(1),1-32.
[17]刘贤敏,刘唱.中国古典音乐诱发情绪的生理活动研究[D].南京师范大学.2006,5.
[18]Wagner J, Kim J, Andre E. From Physiological Signals to Emotions: Implementing and Comparing Selected Methods for Feature Extraction and Classification [A]. In IEEE International Conference on Multimedia & Expo (ICME 2005) [C].2005,940～943.
[19]Lan Li and Ji-hua Chen. Emotion Recognition Using Physiological Signals [A]. ICAT 2006 [C]. LNCS 4282,2006,437-446.
[20]Mi-hee Lee, Gyunghye Yang, Hyoung-Ki Lee and Seokwon Bang. Development Stress monitoring System based on Personal Digital Assistant (PDA) [A]. Proceedings of the 26th Annual International Conference of the IEEE EMBS San Francisco [C]. CA, USA,2004,2364-2367.
[21]Timo Partala, Veikko Surakka. Pupil size variation as an indication of affective processing[J]. International Journal of Human Computer Studies.2003, 59(1),185～198.
[22]Jing Zhai and Armando Barreto. Stress Detection in Computer Users Based on Digital Signal Processing of Noninvasive Physiological Variables [A]. Proceedings of the 28th IEEE EMBS Annual International Conference [C]. New York, USA.2006,1355～1358.
[23]Julio C. D. Conway, Claudionor J. N. Coelho Jr., et al. Wearable Computer as a Multi-parametric Monitor for Physiological Signals [J]. IEEE 2000,236～242.
[24]Healey J. A. Wearable and Automotive Systems for Affect Recognition from Physiology. [PhD thesis] MIT, Cambridge, MA,2000.
[25]韩伟,张泾周,刘园园,张光磊.脑电节律提取的小波变换方法[J].中国医疗器械杂志,2006,30(1),19～21.
[26]葛哲学沙威.小波分析理论与MATLAB R2007实现[M].北京：电子工业出版社,2007.10.
[27]C.Sidney Burrus等著.程正兴译.小波与小波变换导论／(美)[M].北京：机械工业出版社,2007.10.
[28]Glover F., Future paths for integer programming and links to artificial intelligence, Computers and Operations Research,1986,13,533～549.
[29]Glover F. and M.Tabu Search. Boston, Kluwer Academic Publishers,1997.
[30]邢文训,谢金星.现代优化计算方法[M].清华大学出版社.2005,51-68.
[31]王凌.智能优化算法及其应用[M].北京：清华大学出版社,施普林格出版社.2010.10.
[32]Randall S. Sexton, Bahrain Alidaee, Robert E. Dorsey, and John D.Johnson, Global Optimization for Artificial Neural Networks:A Tabu Search Application, European Journal of Operational Research,1998,106, pp,570～584.
[33]Hughes J R. EEG in clinical practice. Butterworth-Heinemann,1994.
[34]Shi Y., Eberhart R C.A modified swarm optimizer [A].IEEE International Conference of Evolutionary Computation [C]. Anchorage, Alaska:IEEE Press, 1998.
[35]边肇祺,张学工.模式识别[M].清华大学出版社.2000,176-210.
[36]Sergios Theodoridis and Konstantinos Koutroumbas. Pattern Recognition, Second Edition,2004,106～124.