基于判决理论的在线BCI系统的研究与建立

发布时间：2018-03-05 18:25

本文选题：脑-机接口　切入点：序贯概率比检验　出处：《大连理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：腑－机接口(Brain-computer Interface, BCI)是不依赖神经肌肉组织的信息交流通道,它为患有运动障碍的病人与外界交互提供了一种新的手段。BCI的出现最初是以临床应用为目的,但近年来也已开始向非临床应用领域发展,所以BCI系统的在线实现具有十分重要的研究意义和应用价值。 BCI系统的建立包括离线分析和在线实现两部分。目前虽然有许多关于脑电信号分析方法的介绍,并在竞赛数据中取得很高的分类正确率,但有许多方法因运算速度、算法复杂度等原因并不具有实用性,因此算法的实时性是确保在线实现的关键。本文着眼于建立基于左右于运动想象的BCI系统,并在提供给受试者实时反馈的情况下实现光标的方向控制,基于此本文分别使用自回归(Autoregression, AR)模型和自适应Morlet小波基作为特征提取方法,结合根据判决理论建立序贯假设检验(Sequence Probability Ratio Test, SPRT)和序贯判别分析(Sequential Linear Discriminant Analysis, SLDA)两种分类器,对国际BCI竞赛提供的数据集进行离线分析,验证算法识别准确率,实现动态分类。竞赛数据的仿真分析结果表明：采用自适应Morlet小波基提取特征比AR模型法能够取得更高的分类准确率,在同样的特征提取方法下,两种分类器对竞赛数据的分类效果相当。为验证算法的实时性能及实用性,本文基于BCI2000软件平台,以事件相关去同步、同步为基础,设计了左右手运动想象实验,该实验包含校准实验以及光标控制实验两个部分。离线分析中,首先使用BCI2000的离线分析工具对每位受试者定位两种任务类型下最具区分度的导联位置,然后采用自适应小波基进行特征提取,在4位受试者初次的实验数据中,SPRT获得了72.88%的平均正确率,SLDA获得76.21%的平均准确率,且SLDA所需判决时间要短于SPRT。基于离线分析结果,选取两名准确率最高的受试参与光标控制实验,采用自适应小波基结合SLDA作为实时算法,经训练后两名受试的平准正确率均高于80%。该研究为国内在线运动想象BCI系统的一个重要突破,也为今后的BCI系统的实时性及实用化研究积累了经验。
[Abstract]:Brain-computer Interface (BCI) is a channel of information exchange independent of neuromuscular tissue, which provides a new means for patients with motor disorders to interact with the outside world. BCI was originally used for clinical purposes. However, in recent years, the field of non-clinical application has begun to develop, so the online implementation of BCI system has very important research significance and application value. The establishment of BCI system consists of offline analysis and online implementation. Although there are many introductions on EEG analysis methods and high classification accuracy in race data, there are many methods due to the speed of operation. The complexity of the algorithm is not practical, so the real-time of the algorithm is the key to ensure the online implementation. This paper focuses on the establishment of a BCI system based on the left and right motion imagination. In this paper, autoregressive autoregressive (ARG) model and adaptive Morlet wavelet basis are used as feature extraction methods. Two classifiers, Sequence Probability Ratio Test (SPRT) and Sequential Linear Discriminant Analysis (SLDAs), are established according to the decision theory. The data set provided by the international BCI competition is analyzed offline, and the recognition accuracy of the algorithm is verified. The simulation results of competition data show that the adaptive Morlet wavelet basis method can achieve higher classification accuracy than AR model. The two classifiers have the same classification effect on contest data. In order to verify the real-time performance and practicability of the algorithm, based on the BCI2000 software platform, based on the event related de-synchronization and synchronization, this paper designs the left-right hand motion imagination experiment. The experiment consists of two parts: calibration experiment and cursor control experiment. In offline analysis, we first use BCI2000's off-line analysis tool to locate the most differentiated lead position for each of the two task types. Then adaptive wavelet basis is used to extract the features. In the first experiment data of 4 subjects, the average correct rate of 72.88% and the average accuracy of 76.21% are obtained, and the decision time of SLDA is shorter than that of SPRT.Based on the results of off-line analysis, Two subjects with the highest accuracy were selected to participate in the cursor control experiment, and the adaptive wavelet base combined with SLDA was used as the real-time algorithm. After training, the accuracy rate of the two subjects is higher than that of 80%. This research is an important breakthrough in domestic online motion imagination BCI system, and it also accumulates experience for the real-time and practical research of BCI system in the future.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：R318

【参考文献】