基于三维编码刺激序列的视觉P300-Speller诱发ERP研究

发布时间：2018-02-01 09:57

本文关键词： 脑-机接口 P300-Speller事件相关电位支持向量机线性判别分析基于集成学习思想的支持向量机递归特征筛选　出处：《天津大学》2012年硕士论文　论文类型：学位论文

【摘要】：P300-Speller是利用稀少事件相关电位(Event-Related Potential, ERP)—P300信号特征实现文字选择输入的经典人机交互范式。它能够实现受试者利用脑电与外界进行文字交流的功能,已公认是脑-机接口(Brain-Computer Interface, BCI)技术中目前最为有效的信息输入与交互的重要手段之一。传统6×6行列字符闪烁模式的P300-Speller因其存在可选字符数目有限、信息传输效率低、不利于大指令集传输等问题,难以满足实际应用需求。为此,引入新的编码刺激模式以改进传统P300-Speller范式、扩展其字符数、提高其信息传输率并保持较高的分类正确率是研发实用型P300-Speller BCI亟待解决的关键技术。本文首次提出了基于三维编码字符的闪烁刺激模式,并对该模式诱发产生的P300信号特征进行了细致分析,论证了其取得较高分类正确率和信息传输率的可行性。在此基础上研究设计了基于三维编码闪烁模式的64字符和125字符两种改进型P300-Speller;完成了传统6×6行列字符、三维编码64字符和125字符三种刺激模式实验;对实验数据进行了具体的预处理、特征提取与模式识别及分类正确率与信息传输率的比较分析。研究中,首先对实验数据进行滤噪、降采样等预处理。之后,分别采用Fisher系数和r2系数方法对脑电特征进行了可分性分析;利用相干平均方法提取了脑电特征并分别采用线性判别分析(Linear Discriminant Analysis, LDA)和支持向量机(Support Vector Machine, SVM)方法进行了脑电模式识别,所有被试的平均分类正确率可以达到99%以上。最后,比较了传统行列模式和新型三维编码模式的信息传输速率(Information Transfer Rate, ITR),结果表明三维编码的ITR(最高平均值为53.59bit/min)明显高于行列模式(最高平均值为32.94bit/min)。文中还利用基于集成学习思想的支持向量机递归特征筛选方法进行了导联优化。优化结果表明,大部分受试者的64导联实验数据在去掉60个次要导联后分类正确率还能稳定在80%以上,且保留的重要导联集中在头顶部,与神经电生理学预示以头皮电极监测P300的最佳位置一致。以上研究结果可望为设计开发具有大字符集、大指令集、高信息传输率和高分类正确率的新型实用P300-Speller脑-机接口提供关键技术保障。
[Abstract]:P300-Speller is Event-Related Potential using rare event-related potentials. ERP)-P300 signal features realize the classical human-computer interaction paradigm of text selection input, which can realize the function of using EEG to communicate with the outside world. It has been recognized as Brain-Computer Interface. The traditional 6 脳 6 column character scintillation mode P300-Speller has a limited number of optional characters because of its existence. Information transmission efficiency is low, which is not conducive to the transmission of large instruction sets, so it is difficult to meet the needs of practical applications. Therefore, a new coding stimulation model is introduced to improve the traditional P300-Speller paradigm. Expanding the number of its characters, improving its information transmission rate and maintaining a high classification accuracy are the key technologies to be solved in the development of practical P300-Speller BCI. In this paper, for the first time, a flicker stimulation model based on 3D coded characters is proposed, and the characteristics of P300 signal induced by this model are analyzed in detail. The feasibility of achieving high classification accuracy and information transmission rate is demonstrated. Based on this, two improved P300-Spell models based on 3D encoding scintillation mode, 64 characters and 125 characters, are studied and designed. Er; The experiments of traditional 6 脳 6 column character, 64 character 3D coding and 125 character stimulation mode are carried out. The experimental data are preprocessed, feature extraction and pattern recognition, classification accuracy and information transmission rate are compared and analyzed. In the study, the experimental data were preprocessed with noise filtering and sampling reduction. Then, the Fisher coefficient and R2 coefficient were used to analyze the separability of EEG characteristics. The EEG features were extracted by the coherent averaging method and linear Discriminant Analysis was used respectively. LDAs and support vector machine support Vector machines (SVM) were used to recognize EEG patterns. The average classification accuracy of all subjects can reach more than 99%. Finally. The information Transfer rate (ITR) of the traditional row-column mode and the new 3D coding mode are compared. The results show that the ITR (maximum average value is 53.59 bit / min) in 3D coding is significantly higher than that in column model (highest average value is 32.94 bit / min). In this paper, the support vector machine recursive feature selection method based on integrated learning is also used to optimize the lead. The optimization results show that. Most of the 64 lead data were stable above 80% after the 60 secondary leads were removed, and the remaining important leads were concentrated on the top of the head. These results are consistent with the best position predicted by neuroelectrophysiology to monitor P300 with scalp electrode. The above results are expected to have a large character set and a large instruction set for design and development. The new practical P300-Speller brain-computer interface with high information transmission rate and high classification accuracy provides key technical support.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R318.0

【参考文献】