【作者】 安兴伟；

【导师】 万柏坤；

【作者基本信息】 天津大学 ， 生物医学工程， 2010， 硕士

【摘要】 与感觉、认知密切相关的事件相关电位(Event Related Potential,ERP)是是目前脑-机接口(Brain Computer Interface, BCI)系统中较为常用的提取大脑思维特征信息载体,其中主要内源性成分为P300。但长期以来ERP的可识别性与诱发速度间存在着相互制约的瓶颈,影响着BCI信号特征提取准确率与信息转化速度。为解决这一难题,本文引入了一种新型的视-听联合刺激模式替代传统的单一视觉刺激模式以诱发产生跨感觉通路的ERP信号,并研究了该信号特征,探讨了兼顾提高识别效率和传输速度的可能途径。本课题在充分调研ERP信号产生和P300成分特点及刺激诱发ERP现状的基础上,设计了基于视-听联合刺激选择数字的跨感觉诱发ERP实验方案,对传统单一视觉诱发P300 speller刺激范式做了实质性的改进,在视觉数字Oddball序列(简化为1～9数字)中加入了相应数字语义的汉语声音刺激以实施跨感觉交叉刺激的脑-机交互任务。本文使用NeuroScan4.3数字脑电采集(64导)系统分别进行了经典的单一视觉刺激范式和上述视-听联合刺激范式的两组实验,并对其诱发产生ERP信号和P300成分特征与识别方法做了较细致的比较分析。研究中首先将经实验得到的脑电信号数据采用神经电生理学常规分析方法进行处理。结果显示,视-听联合刺激范式诱发的P300成份较单一视觉刺激范式具有更高的波幅和更短的潜伏期,与文献报道吻合,表明跨感觉通路刺激模式更适于高速脑-机信息交互。本文进而采用新的约束独立分量分析(constrained Independent Component Analysis, cICA)方法对视-听交叉刺激诱发的ERP信号进行了特征提取,获得了与实验前预设参考信号关系最密切的非高斯分量特征;其后以部分导联的P300降采样波形为基本特征,再结合前述cICA提取的分量特征,最后采用支持向量机分类算法进行了脑-机交互任务识别。对多次重复刺激情况下的靶刺激进行识别时以集成判别函数的多重输出作为最终判决依据。结果表明,视-听联合刺激较单一视觉刺激诱发的ERP信号特征更强,不但有更好的识别效率,且可在较高识别率情况下降低所需重复刺激次数以缓解识别效率和传输速度相互制约的困难;采用跨感觉通路联合刺激作为脑-机交互范式有利于摆脱单一通路模式的困境、有效地提高脑-机信息交互能力,值得进一步深入研究。更多还原

【Abstract】 Event Related Potential (ERP) which closely correlated with the sense perceptive cognitions is an important neurophysiology paradigm for Brain Computer Interface (BCI). And in ERP, the P300 component is one of the most widely used endogenous components for BCI studies. Unfortunately, there is a difficulty of restricting condition each other between ERP recognition and evoking rate, which affects the accuracy of ERP feature extraction and its information transmitting rate badly. To try to improve the rate of ERP information transmitting and the accuracy of ERP pattern classification simultaneously, in this thesis, the visual-auditory evoking model was used instead of visual evoking model and the features of the signals were analyzed.Based on thorough investigation about the generation of ERP、the feature of P300 components and the present situation of ERP stimulation, the traditional P300 speller stimulating pattern was improved in this thesis, which was a cross model evoking experiment for brain computer interface providing corresponding semantic auditory stimulating simultaneously when visual stimulation (number 1-9) is proved. The NeuroScan 4.3 digital sampled system was used to acquire the signals we need. The subjects were divided into two groups, the first group used the traditional P300 speller pattern, while, the second group used the new visual-auditory stimulation pattern we designed.The EEG signals acquired in this system were analyzed by conventional neurophysiological analysis. Results showed that the P300 components in visual-auditory cross model processing had higher amplitudes and shorter latency compared with unimodal visual evoking, which was consistent with existing research and proves that the cross model pattern was more suitable for high-speed brain machine interface. The constrained Independent Component Analysis (cICA)was used in feature extracting to extract the most non-Gaussian components which was closely related to the reference signal. Then the resampled P300 features of some important channels combined with the independent component extracted by cICA were used as the feature vector for recognition tasks. The multiple output decision values of support vector machine were applied to build the classification algorithm, which was used to identify the target number.Results showed that visual-auditory cross stimulation can evoke better EEG feature, which can lead to a higher classification rate and lower the number of the repeated trials. Thus, the accuracy of identify the target number and its information transmitting rate can be improved simultaneously. Studies showed that using the visual-auditory cross stimulation as the model of the brain machine interface can solve the difficulty existed in unimodal stimulation and effectively improve the ability of brain machine interface.更多还原