基于事件相关电位的跨个体脑—机接口研究

发布时间：2018-04-25 06:56

本文选题：脑-机接口 + 事件相关电位　；参考：《天津大学》2014年硕士论文

【摘要】：与传统的个体依赖型脑-机接口相比,跨个体脑-机接口的用户可以缩短或者不经历校准过程,直接对通用型系统进行使用。缩短脑-机接口校准过程意味着对肢体运动功能受损患者时间精力的节省和后续使用效果的提高,同时也更加符合面向大量健康用户的通用型脑-机接口的使用需求。然而目前已有的基于事件相关电位(ERP)的脑-机接口跨个体识别研究大多存在被试样本数少,缺少实时验证等问题,其可行性及其影响因素需要进一步的探究。针对上述问题,本文设计了两部分实验,在离线实验中利用Farwell范式视觉诱发ERP信号,采集了55名被试数据,利用相干平均算法提取ERP时域特征,并通过一种集成的线性分类器进行跨个体的字符识别,得到86.64%的平均正确率,充分显示了脑-机接口跨个体识别的可行性,其中65.45%的被试跨个体识别率达到90%,表明对大部分被试来说,系统使用前的校准过程甚至可以省略。研究中进一步对影响跨个体识别效果的因素进行了分析,其中加入被试自身信息可以明显提升跨个体模型的识别效果,相比单纯利用其自身数据建模而言,平均时间节省约93.93%,大大加快了校准进程。另外,实验结果也显示训练跨个体模型的人数以及性别等因素会对其识别效果造成影响。为了进一步验证离线分析的结论,对20名被试在相同的实验范式下进行了在线测试,利用其一周前的离线数据所建立的SD,SI,Semi SI三种模型进行字符实时预测,分别得到了97.5%,87.7%和91.1%的识别正确率,验证了同一个体不同时间的脑电信号稳定性,以及跨个体分类器在线使用的可行性。为了探索跨个体模型的可推广性,文中还在两组不同实验范式的数据集上进行了识别验证,测试了该模型的适用范围和通用性能。总之,本文研究成果从个体、时间、实验范式等多个角度,探索了跨个体通用模型的可行性,为通用型脑-机接口应用的进一步发展提供了理论与技术支持。
[Abstract]:Compared with the traditional individual dependent brain-computer interface, the cross-individual BCI user can shorten or not go through the calibration process and directly use the general-purpose system. Shortening the calibration process of BCI means saving time and energy and improving the effect of subsequent use in patients with impaired limb motor function. At the same time, it is more suitable for the use of general brain-computer interface (BCI) for a large number of healthy users. However, most of the existing researches on BCI based on ERP (event related potentials) have few samples and lack of real time verification, so the feasibility and its influencing factors need to be further explored. In order to solve the above problems, two experiments are designed in this paper. In the off-line experiment, the Farwell normal form is used to induce the ERP signal, 55 subjects are collected, and the time domain feature of ERP is extracted by using the coherent averaging algorithm. An integrated linear classifier is used to recognize characters across individuals, and an average accuracy of 86.64% is obtained, which fully shows the feasibility of BCI cross-individual recognition. Among them, 65.45% of the subjects had a cross-individual recognition rate of 90%, indicating that the calibration process before the use of the system could even be omitted for most of the subjects. In the study, the factors that affect the cross-individual identification effect are analyzed, and the identification effect of cross-individual model can be significantly improved by adding the information of the subjects, compared with the mere use of their own data to model the model. Average time savings are about 93.93, greatly speeding up the calibration process. In addition, the experimental results also show that the number and gender of training cross-individual model will affect the recognition effect. In order to further verify the conclusion of offline analysis, 20 subjects were tested on line under the same experimental paradigm, and the three models of SDFSI Semi SI, which were established by their offline data a week ago, were used to predict the characters in real time. The recognition accuracy of 97.5% and 91.1% is obtained respectively, which verifies the stability of EEG signals at different time points of the same body and the feasibility of on-line use of cross-individual classifiers. In order to explore the extensibility of the cross-individual model, two sets of data sets of different experimental paradigms are used to identify and verify the model, and the applicability and general performance of the model are tested. In a word, from the perspectives of individual, time and experimental paradigm, this paper explores the feasibility of cross-individual general model, which provides theoretical and technical support for the further development of general-purpose brain-computer interface applications.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R318;TN911.7

【相似文献】