【作者】 于波；

【导师】 李海峰；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2017， 博士

【摘要】 认知控制(Cognitive control)是指人们执行任务中面对干扰或冲突时选择任务相关信息的加工,抑制不相关信息的加工,从而灵活地完成当前任务的能力。认知控制是我们人类大脑非常重要的一种执行能力,它与脑部疾病、情绪疾病、精神疲惫的发生密切相关。目前人脑认知控制的研究存在的主要问题有:认知控制实验范式类别不足、认知控制机制及模型不清楚以及认知控制能力相关的计算指标的缺陷,缺少实时认知控制检测识别方法,从而导致缺少认知控制能力评价的方法。本文主要的研究目的是建立认知控制评测实验系统,通过采集实验者进行认知控制任务时产生的脑电数据,研究和探索人脑认知控制的认知规律及相应机制,进而提出认知控制能力评价相关的生物计算指标,并基于认知控制的认知规律和生物计算指标提出从单次试验脑电数据中提取认知控制相关的特征,最后,根据样本特征空间的分布信息构建认知控制的模式分类方法,提高认知控制脑电的检测识别率。本论文针对人脑认知控制计算指标及其检测识别方法等相关问题和挑战进行了深入的研究,主要的研究工作和创新成果包括以下五个方面:第一,针对人脑认知控制计算困难及评估的问题,提出了人脑认知控制计算指标及其检测识别方法的整体解决方案。首先,以认知控制的研究理论为基础,系统地提出认知控制相关的生物计算指标,进而提出认知控制相关的模型。其次,改进和完善认知控制的实验范式,设计认知控制的实验平台,包括:非注意状态下非听觉通道(复杂信息)认知控制实验范式和前注意阶段基于汉语语料的听觉认知控制实验范式。然后,对采集得到的实验者的脑电数据进行处理分析,提出计算认知控制能力评价相关的生物计算指标方法,并提出人脑认知控制处理模型。最后,基于上述研究,从脑电特征提取方法和模式识别方法两个方面,提出解决针对单次试验获得的脑电数据进行认知控制的检测识别问题的方法,从而为实时地、准确地、在线地对人脑认知控制能力进行评估提供手段。第二,针对在非注意情况下,人脑认知控制是否对周围的非听觉通道(情感通道)的复杂信息存在认知控制加工的问题,提出了非注意状态下情感信息(UAEI,Unattended emontional information)认知控制加工相关的指标的计算方案。首先,采用语义面孔作为刺激材料,设计了UAEI认知控制的实验范式。其次,针对采集得到的实验者的认知脑电信号(Electroencephalogram,EEG),提出基于UAEI-ERD/ERS(Unattended emontional information-Evented-Related Desynchronization/Synchronization,非注意状态下情感信息的事件相关去同步化和同步化)方法计算获得认知控制相关脑波振荡指标。然后,每个实验者的每个时间段的对应电极的平均UAEI-ERD值通过重复测量方差统计分析。实验结果发现更多与UAEI认知控制相关的生物计算指标。从而,首次建立了人脑认知控制系统对UAEI的计算指标——与高兴表情加工有关的β1HU AEI-ERD(UAEI-ERD ofβ1 for happy expression)和与悲伤表情加工有关的β2SU AEI-ERD(UAEI-ERD ofβ2for sad expression)。该研究可用于解决人脑UAEI认知控制的评测问题,并为后续研究人脑听觉通道认知控制工作的开展提供了研究基础和相关研究结论。第三,针对人脑注意状态听觉冲突信息(AACI,Attentive auditory conflicting information)认知控制缺少计算指标和认知模型的问题,提出AACI的认知控制检测的计算指标研究方法。首先,针对听觉认知控制相关研究较少,尤其是采用汉语语料作为实验刺激的实验范式较少的问题,采用Stroop实验范式,设计AACI认知控制的实验范式。其次,对采集得到的脑电进行预处理之后,提出了采用AACI-ERP方法计算听觉认知控制脑电信号相关的计算指标。之后,提出AACI认知控制相关指标的计算指标,包括:感知计算指标(Sensory computation indices,SCI)、确认计算指标(Identification computation indices,ICI)和执行计算指标(Execution computation indices,ECI)。最后,通过基于短时间窗的AACI-ERP计算指标的分析方法证明本文提出的计算指标的有效性。相比于其他研究方法获得的计算指标,实验结果发现一系列显著的AACI-ERP计算指标,揭示了一致和不一致试验条件下认知控制的差异。基于研究得到计算指标,提出更详细具体的三阶段认知控制模型,用于描述人脑在前注意阶段听觉认知控制的机制。该研究完善了听觉认知控制的评测方法。第四,针对缺少从单次试验脑电信号里认知控制相关特征提取方法的问题,提出了听觉认知控制相关的脑电样本特征提取方法。首先,基于认知控制的相关研究,提出了认知控制及认知控制缺失症的检测识别的问题。其次,利用冲突和非冲突的语音信号刺激,分析研究脑电信号,给出听觉认知控制时域的认知规律相关的计算指标。根据AACI相关的计算指标(SCI、ICI、ECI)及三阶段认知控制模型,单次试验脑电样本被分成3个部分。被分割的每个阶段使用时域上的平均幅值和Lempel-Ziv复杂度(LZC)进行计算,从而联合三个阶段的特征作为听觉认知脑电样本的特征。最后通过实验证明了,针对单次听觉脑电样本的特征提取方法(Computing the feature of auditory single trial EEG,CFAST),联合使用平均幅度和LZC可以获得最好的识别率(99.33%)。从而,提出的方法能够有效地检测听觉认知控制脑电数据,进而提供人脑认知控制能力实时评价的检测识别方法。第五,针对缺少对认知脑电信号进行认知控制模式分类的有效方法的问题,提出了考虑样本空间分布信息的支持向量机模式分类方法。首先,研究支持向量机(Support Vector Machine,SVM)在处理样本不平衡分布时存在有偏向性问题,使少数类别的分类错误率的上界高于多数样本类别。其次,分析总结了针对该问题当前的研究方法,并指出存在问题。提出基于样本空间分布信息的SVM分类识别方法。该方法主要是使少数类样本点到超平面的距离大于多数类,降低少数类的错分类的上界,充分利用样本的空间分布距离信息,没有增加或减少样本点。最后,UCI数据集和认知控制脑电集的实验结果显示了三种提出的方法中,MSEDR-SVM(Mean Sample Euclidean Distance Ratio-SVM)方法在处理不平衡样本分布时,是最有效的,可以有效检测识别认知控制脑电,从而为认知控制的实时检测提供模式分类方法。更多还原

【Abstract】 When dealing with the contradiction and distraction during a certain task,human brain has the capabilities to respond flexibly,and one of such capabilities is typically called cognitive control.For the task,cognitive control processes the related information and suppresses irrelevant cues.In this sense,cognitive control is one of the most important capability of human brain,and it is closely related with brain disease,emotional disorder and neurolysis.Although the research on cognitive control has quite a long history,to the best of the author’s knowledge,there are few feasible methods for detecting and evaluating the capability of cognitive control,which is mainly caused by the following reasons: the cognitive control experiments are confined to quite limited number of paradigms,the cognitive control mechanism and the relevant computation indices are not well established,and additionally,there is no real-time approach to cognitive control detection.To establish the experiment system for evaluating cognitive control,this dissertation collects the EEG data produced during cognitive control tasks to explore the rules and mechanisms of cognitive control.On this basis,this dissertation proposes computation indices for cognitive control capability evaluation,extracted cognitive control-related feature from each single trial EEG,and constructs cognitive control classification method according to the feature space distribution of EEG samples.This dissertation focuses on the computation indices and detection of cognitive control,and the main contribution are as follows:Firstly,computation indices and the corresponding detection method for cognitive control were proposed.Based on the cognitive theory,this dissertation systematically presented biological indices for cognitive control,and proposed a cognitive control model.Additionally,the existing paradigms,which are a cognitive control experiment for non-acoustical modality under unattended situations and an auditory cognitive control paradigm under preattentive stage based on the Chinese speech stimuli,were improved,and the relevant testbed for cognitive control was designed.Through analyzing the EEG data,the method for calculating the indices of cognitive control capability evaluation was proposed,and a model for the cognitive control was established.Besides,on the basis of the previous research,this dissertation proposed a method which can detect and recognize cognitive control from a single trial EEG,so that online evaluation of cognitive control is enabled.Secondly,to solve the problem that the cognitive control mechanism triggers spontaneous processing of non-acoustical complicated information under unattended condition,this dissertation proposed the computation scheme for the calculation of cognitive control for unattended emontional information(UAEI).Using the semantic faces as stimuli,we designed a paradigm for UAEI cognitive control.Using the UAEI-ERD/ERS to calculate the brain oscillations indices related to cognitive control.Then,the average UAEI-ERD values across the respective electrode sites for each time interval were entered into repeated measures analysis of variance(ANOVA).The experimental results indicated more computation indices for UAEI cognitive control.The current work firstly constructs UAEI computation indices——β1HU AEI-ERD(UAEI-ERD of β1 for happy expression)and β2SU AEI-ERD(UAEI-ERD of β2 for sad expression).The study solves the problem of UAEI cognitive control evaluation,and provides conclusions and basis for future cognitive control of acoustic channel.Thirdly,this dissertation proposed the research method for computing cognitive control quantitative indices for attentive auditory conflicting information(AACI),which are missing in the previous research.Both of existing work pay little attention to acoustical cognitive control,especially to the experimental paradigms using Chinese stimuli,and such situation motivates the authors to design,using the Stroop paradigm,we designed a paradigm for AACI cognitive control.As well as AACI-ERP(AACI-event-related potential)processing method to process the sampled and preprocessed EEG.Then we proposed the method of computing indices of AACI cognitive control,including: SCI(Sensory computation indices),ICI(Identification computation indices),ECI(Execution computation indices).A smaller AACI-ERP statistical method evidenced the validity of the proposed indices.The experimental results presented in this dissertation lead to a series of significant quantitative indices,and revealed the difference between the cognitive control under the consistent and inconsistent conditions.Finally,based on the the computation indices,this dissertation proposed an elaborate 3-stage cognitive control model which depicts the mechanism of AACI cognitive control.The study improves the evaluation methods of auditory cognitive control.Fourthly,this dissertation proposed a feature extraction method,which is able to extract features related to cognitive control from a single trial EEG data.Based on the research on cognitive control,a cognitive control disorder detection problem is presented.Then through the stimulation caused by both the conflicting and non-conflicting speech,this dissertation analyzed EEG signal and proposed the computation indices related to temporal cognitive rules of acoustical cognitive control.Additionally,this dissertation studied the feature extraction for single trial EEG sample based on cognitive control mechanism of speech confliction.Based on the computation indices relevant to AACI and 3-stage cognitive control model,each EEG sample for a single trial was divided into three parts.Average amplitude in the time-domain and Lempel-Ziv Complexity(LZC)was computed for the divided part of each EEG sample and the union of the feature of the three stages was used as the feature of each auditory cognitive EEG sample(Computing the feature of auditory single trial EEG,CFAST).The experimental results indicated that,for the single trial auditory cognitive control feature extraction,the combination of average amplitude and LZC can achieve the best accuracy(99.33%),hence the proposed method can effectively detect auditory cognitive control in single trial EGG,which enables real-time evaluation of brain cognitive control.Fifthly,a SVM pattern classifier,which take into consideration the sample distribution,is proposed to adapt the cognitive control classification problem of EEG signal.It is well known that traditional SVM suffers from bias when dealing with unevenly distributed samples,which raises the upper bound of error rate for some class.To solve the previous problem,this dissertation proposed a SVM classifier based on the sample distribution.The proposed classifier does not increase or reduce samples,and the key idea is to make the members of the minority classes farther from the hyperplane than those of the majority classes to lower the upper bound of error rate.The experimental results conducted on the UCI dataset and cognitive control EEG dataset demonstrated that the proposed classifier can effectively classify the cognitive control EEG signal.更多还原