无模型贝叶斯分类器方法及在脑电波信号识别中的应用

发布时间：2018-11-01 20:00

【摘要】：随着科技的发展,对于那些大脑健康而肢体行动不便或者无法行动的患者,例如:肌萎缩侧索硬化症(ALS)患者,越发的渴望能够通过自己的大脑直接与外部设备进行交流,脑-机接口(Brain-Computer Interface,BCI)技术就在这种背景下出现并快速发展。脑电波信号的分类算法作为BCI技术中一个重要的环节,对其进行研究将有着重大的现实意义及实用价值。本文的主要研究内容:1.针对朴素贝叶斯分类器(Naive Bayesian Classifier, NBC)及贝叶斯网络(Bayesian Network, BN)的不足进行了改进,基于最近邻(Nearest Neighbor, NN)算法,建立一个计算联合概率分布的概率估计器,并提出一种无模型贝叶斯分类器(Model-Free Bayesian Classifier, MFBC)算法实现数据的分类。2.提出一种新型的自组织特征提取算法,并对MFBC算法进行扩展,从而能够处理回归问题,通过与极限学习机(Extreme Learning Machine,ELM)进行比较,从而验证了扩展后的MFBC处理回归问题的有效性。3.根据模糊C均值(Fuzzy C Means, FCM)聚类算法对人造脑电波信号数据与真实脑电波信号数据进行区分。4.进行对脑电波信号识别系统的设计,包括UCI (University of California Irvine)数据及脑电波数据的导入、MFBC算法、实验结果分析等功能。最后通过UCI数据进行MFBC算法的敏感性分析,同时证明MFBC算法的收敛性。通过与其他经典的及集成分类器进行比较,验证了 MFBC算法的有效性。同时将MFBC算法与公共空间模型(Common Spatial Pattern,CSP)算法相结合应用于BCI系统中的信号分类环节中,从而证明其实用性。
[Abstract]:With the development of science and technology, for those with healthy brain, physically disabled or unable to move, such as amyotrophic lateral sclerosis (ALS), there is a growing desire to communicate directly with external devices through their own brains. Brain-computer interface (Brain-Computer Interface,BCI) technology emerged in this context and developed rapidly. As an important part of BCI technology, the classification algorithm of brain wave signal will have great practical significance and practical value. The main contents of this paper are as follows: 1. Aiming at the shortcomings of naive Bayesian classifier (Naive Bayesian Classifier, NBC) and Bayesian network (Bayesian Network, BN), a probabilistic estimator to calculate the joint probability distribution is established based on the nearest neighbor (Nearest Neighbor, NN) algorithm. A modelless Bayesian classifier (Model-Free Bayesian Classifier, MFBC) is proposed to classify data. 2. In this paper, a new self-organizing feature extraction algorithm is proposed, and the MFBC algorithm is extended to deal with the regression problem, which is compared with the extreme learning machine (Extreme Learning Machine,ELM). The effectiveness of the extended MFBC in dealing with regression problems is verified. 3. 3. According to the fuzzy C-means (Fuzzy C Means, FCM) clustering algorithm, the artificial EEG signal data and the real EEG signal data are distinguished. 4. The design of brain wave signal recognition system, including UCI (University of California Irvine) data and brain wave data import, MFBC algorithm, experimental results analysis and other functions. Finally, the sensitivity of MFBC algorithm is analyzed by UCI data, and the convergence of MFBC algorithm is proved. Compared with other classical and integrated classifiers, the effectiveness of MFBC algorithm is verified. At the same time, the MFBC algorithm and the common space model (Common Spatial Pattern,CSP) algorithm are applied to the signal classification in the BCI system, which proves its practicability.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R318;TN911.7

【相似文献】