基于P300的脑机接口及其在线半监督学习
发布时间:2018-04-26 15:08
本文选题:P300脑机接口 + 字符输入 ; 参考:《华南理工大学》2014年硕士论文
【摘要】:在人工智能、模式识别和信号处理等领域,脑机接口(Brain-computer interface,BCI)技术逐渐成为备受关注的课题,并对失去肢体运动能力的残疾人产生了突出作用。目前,该技术虽然逐渐应用于医学和娱乐等各个领域,但仍存在系统通讯速率较低、训练时间较长等问题。针对上述问题,本文主要做了以下两个方面的研究工作: 首先,本文设计了一种改进的基于P300的字符输入BCI系统,该系统采用的方法是根据受试者的实时脑电数据特征,计算出可能的目标与非目标字符,据此屏蔽这些非目标字符的闪烁,从而缩短闪烁字符序列,并减弱非目标字符闪烁对受试者的干扰。通过采集多位受试者的脑电数据进行分析,实验结果表明,该方法在保持准确率基本没有下降的同时,提高了字符输入的速度,从而提高了系统的信息传输率(Information transferrate,ITR),有利于解决系统的实用性问题。 其次,针对训练时间过长的问题,,可以通过半监督学习减少采集有标记样本的时间,我们研究P300BCI系统的在线半监督学习。通常,半监督学习是先采集少量有标记的数据进行初始模型训练,再利用相对较多的未标记数据进行模型更新。而本文采用的方法是在上述初始模型训练之后,再利用在线获取的未标记数据来逐步更新模型。为了确保分类模型的可靠性,需要对未标记的数据进行样本选择,以降低异常数据对模型的干扰。本文采用的样本选择方法是根据分类模型计算出每个未标记数据的分类器响应值来判定此数据的可信度,即筛选出有利于模型更新的未标记数据来逐步优化分类模型。实验中对在线半监督学习的BCI系统在使用样本选择方法前后进行比较,并分别对采集到的在线数据进行离线分析。实验结果表明,在在线半监督学习的P300BCI系统中使用样本选择方法,不仅能够减少训练时间,而且能够在有标记样本较少的情况下提高分类准确率。
[Abstract]:In the fields of artificial intelligence, pattern recognition and signal processing, Brain-Computer Interface (BCI) technology has gradually become a subject of great concern, and has played an important role in the disabled who have lost the ability of limb movement. At present, although the technology has been gradually applied in various fields such as medicine and entertainment, there are still some problems such as low communication rate and long training time. In view of the above problems, this paper mainly does the following two aspects of research work: Firstly, an improved P300-based character input BCI system is designed. The method of the system is to calculate the possible target and non-target characters according to the real time EEG data characteristics of the subjects. Therefore, the flicker of these non-target characters can be shielded so as to shorten the sequence of scintillation characters and reduce the interference of non-target character flicker to the subjects. By collecting EEG data of many subjects for analysis, the experimental results show that the method can improve the speed of character input while keeping the accuracy rate unchanged. Thus, the information transfer rate of the system is improved and the practical problem of the system is solved. Secondly, aiming at the problem of long training time, we can reduce the time of collecting labeled samples by semi-supervised learning. We study the online semi-supervised learning in P300BCI system. Usually, semi-supervised learning first collects a small amount of labeled data for initial model training, and then uses relatively more unlabeled data to update the model. The method used in this paper is to update the model step by step by using the unlabeled data obtained online after the initial model training. In order to ensure the reliability of the classification model, it is necessary to select samples from unlabeled data to reduce the interference of abnormal data to the model. The sample selection method used in this paper is to calculate the classifier response value of each unlabeled data according to the classification model to determine the reliability of the data, that is, to select the unlabeled data to update the model to optimize the classification model step by step. In the experiment, the online semi-supervised learning BCI system is compared before and after using the sample selection method, and the collected on-line data are analyzed off-line respectively. The experimental results show that using the sample selection method in the online semi-supervised learning P300BCI system can not only reduce the training time, but also improve the classification accuracy when there are fewer labeled samples.
【学位授予单位】:华南理工大学
【学位级别】:硕士
【学位授予年份】:2014
【分类号】:TP334.7;TP18
【参考文献】
相关期刊论文 前5条
1 王行愚;金晶;张宇;王蓓;;脑控:基于脑-机接口的人机融合控制[J];自动化学报;2013年03期
2 刘美春;;脑-机接口系统的类协同式半监督学习[J];科学技术与工程;2013年19期
3 杨帮华,颜国正,严荣国;脑机接口中基于小波包最优基的特征抽取[J];上海交通大学学报;2005年11期
4 葛瑜;刘杨;周宗潭;胡德文;;用于脑-机接口P300实验的支持向量机分类方法[J];计算机工程与设计;2008年11期
5 王志宇;王宏;李一娜;王旭;;小波相关分析在脑-计算机接口系统中的研究[J];仪器仪表学报;2006年04期
本文编号:1806538
本文链接:https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/1806538.html