基于滤波和嵌入式特征选择方法的应用研究
|
摘要
特征选择是模式识别中的关键问题，特征选择按照评价策略，即是否考虑分类器性能将特征选择方法分成基于滤波的特征选择方法和基于嵌入式的特征选择方法。
首先，先介绍了基于滤波的特征选择方法。基于滤波的特征选择方法是一种计算效率比较高的算法，主要采用了与分类任务相关的各种准则，选择最合适的相关准则对特征进行分类。本文采用的滤波的特征选择算法主要有Fisher线性判别准则，最大相关最小冗余度准则，将这两类准则应用到ECoG皮层脑电信号，对该数据中的64个通道进行特征选择。而这两种方法得出的通道的特征排序与共空域子空间的特征提取方法进行结合，结果验证了最大相关最小冗余度准则对皮层脑电信号的通道选择过程有研究意义的。然后分析了基于启发式搜索策略的滤波算法，并且将这些滤波的特征选择算法应用到葡萄酒数据分类中，通过线性分类器分析后，得出这些特征选择算法对葡萄酒数据分类效果有一定的提高。
然后，介绍了基于嵌入式的特征选择方法。基于嵌入式的特征选择方法是将分类器的性能考虑到特征选择的过程中，通过分类器的表现来选择相应的特征。本文采用了支持向量机回归特征消去算法，并且同样应用到EcoG皮层脑电信号中。另一种方法是基于增减特征分量的嵌入式特征选择算法，将这种算法应用到葡萄酒数据分类过程中，实验验证该算法对提高葡萄酒分类有很大的帮助。
最后，将微分进化算法与共空域子空间分解相结合的一种新的嵌入式特征选择算法，并对这种基于这种嵌入式特征选择算法进行了改进。实验验证该方法对皮层脑电信号的通道选择过程中的效果很好，说明该方法有一定的研究意义。
Feature selection is the key problem of pattern recognition, Feature selectionaccording to the evaluation strategies that consider classifier performance will featureselection method based on filtering into feature selection algorithm based on embeddedand feature selection algorithm.
First of all, first introduced the feature selection algorithm based on filtering. Basedon the characteristics of filter algorithm is a kind of computation efficiency of thealgorithm is high, mainly adopts and classification task all kinds of relevant standards, toselect the most appropriate related standards on the characteristics of the classification.This article USES the filtering feature selection algorithm are mainly Fisher lineardiscrimination criterion, the biggest related minimum redundancy standards, will the twotypes of standards applied to ECoG cortical brain electrical signal, the data of 64channels feature selection. And the two methods of the sort that channel characteristicsand the airspace subspace method of feature extraction make combined results show thebiggest related minimum redundancy standards on cortex of eeg channel selectionprocess meaningful research. Then analyzed based on heuristic search strategy filteralgorithm, and the characteristics of the filter algorithm applied to wine dataclassification, through the linear classifier after analysis, it is concluded that these featureselection algorithm for wine data classification effect was improved.
Then, this paper introduces the feature selection algorithm based on embedded.Based on embedded feature selection algorithm is will the capability of classifier toconsider feature selection process, through the classifier performance to choosecorresponding characteristics. This paper used the support vector machine (SVM)regression characteristic elimination algorithm, and also applied to EcoG cortex eegsignals. Another method is based on the characteristics of the weight increase or decreaseof embedded feature selection algorithm, this kind of algorithm is applied to wine in theprocess of data classification and validate it to improve wine classification algorithm hasvery great help.
Finally, the differential evolution algorithm and the airspace subspacedecomposition of combining a new embedded feature selection algorithm, and the basedon this kind of embedded feature selection algorithm was improved. Experimental resultsverify the method of cortex of eeg channel selection process of the effect very good,which shows that this method has a certain research significance.
首先，先介绍了基于滤波的特征选择方法。基于滤波的特征选择方法是一种计算效率比较高的算法，主要采用了与分类任务相关的各种准则，选择最合适的相关准则对特征进行分类。本文采用的滤波的特征选择算法主要有Fisher线性判别准则，最大相关最小冗余度准则，将这两类准则应用到ECoG皮层脑电信号，对该数据中的64个通道进行特征选择。而这两种方法得出的通道的特征排序与共空域子空间的特征提取方法进行结合，结果验证了最大相关最小冗余度准则对皮层脑电信号的通道选择过程有研究意义的。然后分析了基于启发式搜索策略的滤波算法，并且将这些滤波的特征选择算法应用到葡萄酒数据分类中，通过线性分类器分析后，得出这些特征选择算法对葡萄酒数据分类效果有一定的提高。
然后，介绍了基于嵌入式的特征选择方法。基于嵌入式的特征选择方法是将分类器的性能考虑到特征选择的过程中，通过分类器的表现来选择相应的特征。本文采用了支持向量机回归特征消去算法，并且同样应用到EcoG皮层脑电信号中。另一种方法是基于增减特征分量的嵌入式特征选择算法，将这种算法应用到葡萄酒数据分类过程中，实验验证该算法对提高葡萄酒分类有很大的帮助。
最后，将微分进化算法与共空域子空间分解相结合的一种新的嵌入式特征选择算法，并对这种基于这种嵌入式特征选择算法进行了改进。实验验证该方法对皮层脑电信号的通道选择过程中的效果很好，说明该方法有一定的研究意义。
Feature selection is the key problem of pattern recognition, Feature selectionaccording to the evaluation strategies that consider classifier performance will featureselection method based on filtering into feature selection algorithm based on embeddedand feature selection algorithm.
First of all, first introduced the feature selection algorithm based on filtering. Basedon the characteristics of filter algorithm is a kind of computation efficiency of thealgorithm is high, mainly adopts and classification task all kinds of relevant standards, toselect the most appropriate related standards on the characteristics of the classification.This article USES the filtering feature selection algorithm are mainly Fisher lineardiscrimination criterion, the biggest related minimum redundancy standards, will the twotypes of standards applied to ECoG cortical brain electrical signal, the data of 64channels feature selection. And the two methods of the sort that channel characteristicsand the airspace subspace method of feature extraction make combined results show thebiggest related minimum redundancy standards on cortex of eeg channel selectionprocess meaningful research. Then analyzed based on heuristic search strategy filteralgorithm, and the characteristics of the filter algorithm applied to wine dataclassification, through the linear classifier after analysis, it is concluded that these featureselection algorithm for wine data classification effect was improved.
Then, this paper introduces the feature selection algorithm based on embedded.Based on embedded feature selection algorithm is will the capability of classifier toconsider feature selection process, through the classifier performance to choosecorresponding characteristics. This paper used the support vector machine (SVM)regression characteristic elimination algorithm, and also applied to EcoG cortex eegsignals. Another method is based on the characteristics of the weight increase or decreaseof embedded feature selection algorithm, this kind of algorithm is applied to wine in theprocess of data classification and validate it to improve wine classification algorithm hasvery great help.
Finally, the differential evolution algorithm and the airspace subspacedecomposition of combining a new embedded feature selection algorithm, and the basedon this kind of embedded feature selection algorithm was improved. Experimental resultsverify the method of cortex of eeg channel selection process of the effect very good,which shows that this method has a certain research significance.
引文
[1]边肇祺.模式识别[M].第二版.北京:清华大学出版社, 2000: 176-177, 198-204.
[2]西奥多里蒂斯.模式识别[M].李晶皎,译.第三版.北京:电子工业出版社, 2006: 44-46,138-163.
[3]毛勇,周晓波,夏铮,等.特征选择算法的研究综述[J].模式识别与人工智能, 2007, 20(2):211-218.
[4]陈岩,来海峰,王清.基于Filter-Wrapper的两步特征变量提取方法[J].机电工程, 2010,27(4): 67-68.
[5] Inza I, Larranaga P, Blanco R, et al. Filter Versus Wrapper Gene Selection Approaches in DNAMicroarray Domains[J]. Artificial Intelligence in Medicine, 2009, 31(2): 91-103.
[6] Zhou Xiaobo, Wang Xiaodong, Dougherty E R. Nonlinear Probit Gene Classification UsingMutual Information and Wavelet-Based Feature Selection[J]. Biological System, 2008, 12(3):371-386.
[7] Furey T S, Cristianini N, Duffy N, et al. Support Vector Machine Classification and Validation ofCancer Tissue Samples Using Microarray Expression Data[J]. Bioinformatics, 2000, 16(10):906-914.
[8] Shima K, Todoriki M, Suzuki A. SVM-Based Feature Selection of Latent Semantic Features[J].Pattern Recognition Letters, 2008, 25(9): 1051-1057.
[9] Jack L B, Nandi A K. Fault Detection Using Support Vector Machines and Artificial NeuralNetworks, Augmented by Genetic Algorithms[J]. Mechanical System and Signal Processing,2002, 16(2/3): 373-390.
[10] Verikas A, Bacauskiene M. Feature Selection with Neural Networks[J]. Pattern RecognitionLetters, 2002, 23(11): 1323-1335.
[11] Kudo M, Sklansky J. Comparison of Algorithms That Select Features for Pattern Classifiers[J].Pattern Recognition, 2000, 33(1): 25-41.
[12] Dougherty E R. Small Sample Issues for Microarray-Based Classification[J]. Comparative andFunctional Genomics, 2001, 2(1): 28-34.
[13]刘健,钱猛,张维明.基于Fisher线性判别模型的文本特征选择算法[J].国防科技大学学报,2008, 30(5): 135-136.
[14] Han Chuanpeng, Fu huilong, Chris Ding. Feature Selection Based on Mutual Information:Criteria of Max-Dependency, Max-Relevance, and Min-Redundancy[J]. IEEE Transactions onPattern Analysis and Machine Intelligence, 2005, 27(8): 1226-1237.
[15] Han Chuanpeng, Chris Ding, Fu Huilong. Minimum Redundancy Maximum Relevance FeatureSelection[J]. IEEE Intelligent Systems, 2005, 20(6): 70-71.
[16] Gerwin S, Jurgen M. A Pratical Guide to Brain-Computer Interfacing with BCI2000[M].Springer London Dordrecht Heidelberg New York, 2010: 4-5.
[17] Coyle S, Ward T, Markham C, et al. on the Suitability of Near-Infrared(NIR) Systems for NextGeneration Brain Computer Interfaces[J]. Physiol Meas, 2004, 25(4): 815-822.
[18] Laconte S M, Peltier S J, Hu X P. Real-Time FMRI Using Brain-State Classification Hum BrainMapp[J]. Hum Brain Mapp, 2007, 28(10): 1033-1044.
[19] Mellinger J, Schalk G, Braun C, et al. An MEG-Based Brain Computer Interface(BCI)[J].NeuroImage, 2007, 36(3): 581-593.
[20] Ramsey N F, Heuvel M P, Kho K H, et al. Towards Human BCI Applications Based on CognitiveBrain Systems: An Investigation of Neural Signals Recorded from the Dorsolateral PrefrontalCortex[J]. IEEE Trans Neural Syst Rehabil Eng, 2006, 14(2): 214-217.
[21] Wolpaw J R, Birbaumer N, Heetderks W J. Brain-Computer Interface Technology: A Review ofthe First Intermational Meeting[J]. IEEE Transactions Rehabilitation Engineering, 2000, 8(2):164-173.
[22] Furtscheller G P. Event-Related Synchronization (ERS): An Electrophysiological Correlate ofCortical Areas at Rest[J]. Electroencephalography and Clinical Neurophysiology, 1992, 83(1):62-69.
[23] Toro C, Deuschl G, Thatcher R, et al. Even-Related Desynchronization and Movement RelatedCortical Potentials on the ECoG and EEG[J]. Electronencephalography and ClinicalNeurophysiology, 1994, 93(5): 380-389.
[24]刘冲,赵海滨,等.基于频带能量归一化和SVM-RFE的EcoG分类[J].仪器仪表学报, 2011,32(3): 534-539.
[25] Cortez P, Cerdeira A, Almeida F, et al. Modeling Wine Preferences by Data Mining fromPhysicochemical Properties[J]. Decision Support Systems, 2009, 47(4): 547-553.
[26]张学工.关于统计学习理论与支持向量机[J].自动化学报, 2000, 32(1): 32-42.
[27] Isabelle Guyon, Jason Weston, Stephen Barnhill, et al. Gene Selection for Cancer ClassificationUsing Support Vector Machines[J]. Machine Learning, 2002, 46(1-3): 389-422.
[28]游伟,李树涛,谭明奎.基于SVM-RFE-SFS的基因选择方法[J].中国生物医学工程学报,2010, 29(1): 93-95.
[29]甄斌,吴玺宏,刘志敏,等.语音识别和说话人识别中各倒谱分量的相对重要性[J].北京大学学报, 2000, 37(3): 372-374.
[30]克里斯蒂亚尼,沙维-泰勒.支持向量机导论[M].李国军,王猛,曾华军,译.北京:电子工业出版社, 2004: 12-38.
[31] Alex Smola, Bernhard Schoelkopf. A Tutorial on Support Vector Regression[J]. Statistics andComputing, 2004, 14(3): 199-222.
[32]王学仁,温忠粦,编译.应用回归分析[M].重庆:重庆大学出版社, 1989: 32-45.
[33]王惠文,吴载斌,孟洁.偏最小二乘回归的线性与非线性方法[M].北京:国防工业出版社,2006: 40-56.
[34]杨滨,杨晓伟,黄岚,等.自适应迭代最小二乘支持向量机回归算法[J].电子学报, 2010,38(7): 1622-1624.
[35] Leuthardt E, Schalk G, Wolpaw J, et al. A Brain-Computer Interface Using ElectrocorticographicSignals in Humans[J]. Neural Eng, 2004, 1: 63-71.
[36] Lalt N, Hinterberger T, Widman G, et al. Methods Towards Invasive Human Brain ComputerInterface[J]. Advances in Neural Information Processing System(NIPS). 2005, 17: 737-744.
[37] Grainmann B, Huggins J, Levine S, et al. Towards A Direct Brain Interface Based on HumanSubdural Recordings And Wavelet Packet Analysis[J]. IEEE Trans Biomed Eng, 2004, 51:954-962.
[38] Wei Qingguo, Lu Zongwu, Chen Kui, et al. Channel Selection for Optimizing Feature Extractionin an Electrocorticogram-Based Brain-Computer Interface[J]. Journal of ClinicalNeurophysiology, 2010, 27(5): 321-326.
[39]吕俊.基于CSP_BPSO的脑-机接口电极选择[J].华南理工大学学报(自然科学版), 2010,38(10): 7-13.
[40]刘冲,李春胜,赵海滨,等.特征选择算法在ECoG分类中的应用[J].东北大学学报(自然科学版), 2011, 32(5): 658-660.
[41]苏海军,杨煜普,王宇嘉.微分进化算法的研究综述[J].系统工程与电子技术, 2008, 30(9):1793-1796.
[42]赵光权,彭喜元,孙宁.基于混合优化策略的微分进化改进算法[J].电子学报, 2006, 34(12):2402-2403.
[43]赵光权,彭喜元,孙宁.带局部增强算子的微分进化改进算法[J].电子学报, 2007, 35(5):849-850.
[44]唐艳,刘建新,龚安栋. ICA-CSSD的脑机接口分类[J].电子科技大学学报, 2008, 37(3):466-467.
[45]叶柠,孙宇舸,王旭.基于共空间模式和神经元网络的脑-机接口信号的识别[J].东北大学学报(自然科学版), 2010, 31(1): 12-15.
[46]刘琳,魏庆国. CSSD+AAR模型在脑电信号处理中的应用[J].通信技术, 2008, 42(10):207-209.
[47]吕俊,谢胜利.基于运动想象的皮层脑电图频域模式滤波[J].华南理工大学学报(自然科学版), 2008, 36(5): 101-103.
[48]吕俊,谢胜利.脑机-接口中基于ERS/ERD的自适应空间滤波算法[J].电子与信息学报,2009, 31(2): 314-317.
[49]王金甲,李静,张涛,等.二次映射和遗传算法用于鉴别可视化特征提取[J].系统仿真学报,2009, 21(6): 5081-5082.
[50] Wang Y J, Zhang Z G, Li Y. BCI Competition 2003-Data Set IV: An Algorithm Based on CSSDand FDA for Classifying Signal-Trial EEG[J]. IEEE Transactions on Biomedical Engineering,2004, 51(6): 1081-1086.
[51] Blankertz B, Muller K, Krusienski D J, et al. The BCI Competition III: Validating AlternativeApproaches to Actual BCI Problem[J]. IEEE Trans Neural Sys Rehab Eng, 2006, 14(2): 153-159.
[52]吕俊,李胜勇.基于HHT和PSO-C-均值算法的ECoG分类[J].武汉理工大学学报, 2008,30(1): 102-103.
[53] Robert H Kewley, Mark J Embrechts. Data Strip Mining for The Virtual Design ofPharmaceuticals with Neural Work[J]. IEEE Transactions on Neural Networks, 2000, 11(3):668-678.
[2]西奥多里蒂斯.模式识别[M].李晶皎,译.第三版.北京:电子工业出版社, 2006: 44-46,138-163.
[3]毛勇,周晓波,夏铮,等.特征选择算法的研究综述[J].模式识别与人工智能, 2007, 20(2):211-218.
[4]陈岩,来海峰,王清.基于Filter-Wrapper的两步特征变量提取方法[J].机电工程, 2010,27(4): 67-68.
[5] Inza I, Larranaga P, Blanco R, et al. Filter Versus Wrapper Gene Selection Approaches in DNAMicroarray Domains[J]. Artificial Intelligence in Medicine, 2009, 31(2): 91-103.
[6] Zhou Xiaobo, Wang Xiaodong, Dougherty E R. Nonlinear Probit Gene Classification UsingMutual Information and Wavelet-Based Feature Selection[J]. Biological System, 2008, 12(3):371-386.
[7] Furey T S, Cristianini N, Duffy N, et al. Support Vector Machine Classification and Validation ofCancer Tissue Samples Using Microarray Expression Data[J]. Bioinformatics, 2000, 16(10):906-914.
[8] Shima K, Todoriki M, Suzuki A. SVM-Based Feature Selection of Latent Semantic Features[J].Pattern Recognition Letters, 2008, 25(9): 1051-1057.
[9] Jack L B, Nandi A K. Fault Detection Using Support Vector Machines and Artificial NeuralNetworks, Augmented by Genetic Algorithms[J]. Mechanical System and Signal Processing,2002, 16(2/3): 373-390.
[10] Verikas A, Bacauskiene M. Feature Selection with Neural Networks[J]. Pattern RecognitionLetters, 2002, 23(11): 1323-1335.
[11] Kudo M, Sklansky J. Comparison of Algorithms That Select Features for Pattern Classifiers[J].Pattern Recognition, 2000, 33(1): 25-41.
[12] Dougherty E R. Small Sample Issues for Microarray-Based Classification[J]. Comparative andFunctional Genomics, 2001, 2(1): 28-34.
[13]刘健,钱猛,张维明.基于Fisher线性判别模型的文本特征选择算法[J].国防科技大学学报,2008, 30(5): 135-136.
[14] Han Chuanpeng, Fu huilong, Chris Ding. Feature Selection Based on Mutual Information:Criteria of Max-Dependency, Max-Relevance, and Min-Redundancy[J]. IEEE Transactions onPattern Analysis and Machine Intelligence, 2005, 27(8): 1226-1237.
[15] Han Chuanpeng, Chris Ding, Fu Huilong. Minimum Redundancy Maximum Relevance FeatureSelection[J]. IEEE Intelligent Systems, 2005, 20(6): 70-71.
[16] Gerwin S, Jurgen M. A Pratical Guide to Brain-Computer Interfacing with BCI2000[M].Springer London Dordrecht Heidelberg New York, 2010: 4-5.
[17] Coyle S, Ward T, Markham C, et al. on the Suitability of Near-Infrared(NIR) Systems for NextGeneration Brain Computer Interfaces[J]. Physiol Meas, 2004, 25(4): 815-822.
[18] Laconte S M, Peltier S J, Hu X P. Real-Time FMRI Using Brain-State Classification Hum BrainMapp[J]. Hum Brain Mapp, 2007, 28(10): 1033-1044.
[19] Mellinger J, Schalk G, Braun C, et al. An MEG-Based Brain Computer Interface(BCI)[J].NeuroImage, 2007, 36(3): 581-593.
[20] Ramsey N F, Heuvel M P, Kho K H, et al. Towards Human BCI Applications Based on CognitiveBrain Systems: An Investigation of Neural Signals Recorded from the Dorsolateral PrefrontalCortex[J]. IEEE Trans Neural Syst Rehabil Eng, 2006, 14(2): 214-217.
[21] Wolpaw J R, Birbaumer N, Heetderks W J. Brain-Computer Interface Technology: A Review ofthe First Intermational Meeting[J]. IEEE Transactions Rehabilitation Engineering, 2000, 8(2):164-173.
[22] Furtscheller G P. Event-Related Synchronization (ERS): An Electrophysiological Correlate ofCortical Areas at Rest[J]. Electroencephalography and Clinical Neurophysiology, 1992, 83(1):62-69.
[23] Toro C, Deuschl G, Thatcher R, et al. Even-Related Desynchronization and Movement RelatedCortical Potentials on the ECoG and EEG[J]. Electronencephalography and ClinicalNeurophysiology, 1994, 93(5): 380-389.
[24]刘冲,赵海滨,等.基于频带能量归一化和SVM-RFE的EcoG分类[J].仪器仪表学报, 2011,32(3): 534-539.
[25] Cortez P, Cerdeira A, Almeida F, et al. Modeling Wine Preferences by Data Mining fromPhysicochemical Properties[J]. Decision Support Systems, 2009, 47(4): 547-553.
[26]张学工.关于统计学习理论与支持向量机[J].自动化学报, 2000, 32(1): 32-42.
[27] Isabelle Guyon, Jason Weston, Stephen Barnhill, et al. Gene Selection for Cancer ClassificationUsing Support Vector Machines[J]. Machine Learning, 2002, 46(1-3): 389-422.
[28]游伟,李树涛,谭明奎.基于SVM-RFE-SFS的基因选择方法[J].中国生物医学工程学报,2010, 29(1): 93-95.
[29]甄斌,吴玺宏,刘志敏,等.语音识别和说话人识别中各倒谱分量的相对重要性[J].北京大学学报, 2000, 37(3): 372-374.
[30]克里斯蒂亚尼,沙维-泰勒.支持向量机导论[M].李国军,王猛,曾华军,译.北京:电子工业出版社, 2004: 12-38.
[31] Alex Smola, Bernhard Schoelkopf. A Tutorial on Support Vector Regression[J]. Statistics andComputing, 2004, 14(3): 199-222.
[32]王学仁,温忠粦,编译.应用回归分析[M].重庆:重庆大学出版社, 1989: 32-45.
[33]王惠文,吴载斌,孟洁.偏最小二乘回归的线性与非线性方法[M].北京:国防工业出版社,2006: 40-56.
[34]杨滨,杨晓伟,黄岚,等.自适应迭代最小二乘支持向量机回归算法[J].电子学报, 2010,38(7): 1622-1624.
[35] Leuthardt E, Schalk G, Wolpaw J, et al. A Brain-Computer Interface Using ElectrocorticographicSignals in Humans[J]. Neural Eng, 2004, 1: 63-71.
[36] Lalt N, Hinterberger T, Widman G, et al. Methods Towards Invasive Human Brain ComputerInterface[J]. Advances in Neural Information Processing System(NIPS). 2005, 17: 737-744.
[37] Grainmann B, Huggins J, Levine S, et al. Towards A Direct Brain Interface Based on HumanSubdural Recordings And Wavelet Packet Analysis[J]. IEEE Trans Biomed Eng, 2004, 51:954-962.
[38] Wei Qingguo, Lu Zongwu, Chen Kui, et al. Channel Selection for Optimizing Feature Extractionin an Electrocorticogram-Based Brain-Computer Interface[J]. Journal of ClinicalNeurophysiology, 2010, 27(5): 321-326.
[39]吕俊.基于CSP_BPSO的脑-机接口电极选择[J].华南理工大学学报(自然科学版), 2010,38(10): 7-13.
[40]刘冲,李春胜,赵海滨,等.特征选择算法在ECoG分类中的应用[J].东北大学学报(自然科学版), 2011, 32(5): 658-660.
[41]苏海军,杨煜普,王宇嘉.微分进化算法的研究综述[J].系统工程与电子技术, 2008, 30(9):1793-1796.
[42]赵光权,彭喜元,孙宁.基于混合优化策略的微分进化改进算法[J].电子学报, 2006, 34(12):2402-2403.
[43]赵光权,彭喜元,孙宁.带局部增强算子的微分进化改进算法[J].电子学报, 2007, 35(5):849-850.
[44]唐艳,刘建新,龚安栋. ICA-CSSD的脑机接口分类[J].电子科技大学学报, 2008, 37(3):466-467.
[45]叶柠,孙宇舸,王旭.基于共空间模式和神经元网络的脑-机接口信号的识别[J].东北大学学报(自然科学版), 2010, 31(1): 12-15.
[46]刘琳,魏庆国. CSSD+AAR模型在脑电信号处理中的应用[J].通信技术, 2008, 42(10):207-209.
[47]吕俊,谢胜利.基于运动想象的皮层脑电图频域模式滤波[J].华南理工大学学报(自然科学版), 2008, 36(5): 101-103.
[48]吕俊,谢胜利.脑机-接口中基于ERS/ERD的自适应空间滤波算法[J].电子与信息学报,2009, 31(2): 314-317.
[49]王金甲,李静,张涛,等.二次映射和遗传算法用于鉴别可视化特征提取[J].系统仿真学报,2009, 21(6): 5081-5082.
[50] Wang Y J, Zhang Z G, Li Y. BCI Competition 2003-Data Set IV: An Algorithm Based on CSSDand FDA for Classifying Signal-Trial EEG[J]. IEEE Transactions on Biomedical Engineering,2004, 51(6): 1081-1086.
[51] Blankertz B, Muller K, Krusienski D J, et al. The BCI Competition III: Validating AlternativeApproaches to Actual BCI Problem[J]. IEEE Trans Neural Sys Rehab Eng, 2006, 14(2): 153-159.
[52]吕俊,李胜勇.基于HHT和PSO-C-均值算法的ECoG分类[J].武汉理工大学学报, 2008,30(1): 102-103.
[53] Robert H Kewley, Mark J Embrechts. Data Strip Mining for The Virtual Design ofPharmaceuticals with Neural Work[J]. IEEE Transactions on Neural Networks, 2000, 11(3):668-678.