基于MPI并行框架的实时注意力监测脑机接口的研究与应用

发布时间：2018-08-13 15:36

【摘要】：注意力是心理学的一个概念,同时也是脑科学和生物信息学中的一个研究对象。监测人的注意力可以成为人们学习或者工作的一种辅助工具,进而使人的学习和工作效率得到提高,甚至少犯许多严重的错误,例如在学习的过程中,如果有一个监测学生注意力的工具,那么老师可以在学生注意力不集中的时候提醒学生,学生就可以及时更正自己的精神状态。结合计算机科学及生物信息科学,我们可以通过脑机接口来监测人的注意力。用计算机开发各种用途的脑机接口是当前的一个研究热点,但是,目前绝大部分脑机接口的类型都是动作类型,即通过脑机接口来监测大脑在想象做什么样的动作,进而由脑机接口向外部工作发出相同动作的指令,这种脑机接口的典型应用就是机械臂。而关于实时监测注意力这样的精神状态的脑机接口研究比较欠缺,因为注意力涉及到许多心理知识,研究过程比较复杂。而开发一个实时的监测注意力的应用系统更加少,更多的是离线处理的脑机接口,因为脑电波的分析处理十分复杂,中间的每一处理步骤都需要消耗一定的时间,很难做到实时监测。本论文对当前的脑机接口的技术和基础理论进行了深入的分析,并吸收了诸多脑机接口开发的经验和理论知识,自主开发了一个实时监测的脑机接口,并将其应用于监测人在学习中的注意力。该脑机接口采用小波变换对脑电信号进行去噪,然后用基于汉宁窗的FIR算法提取了6个特征值,最后用KNN (k-Nearest Neighbor)算法进行分类。脑机接口需要经过大量的训练才能应用,通常一个脑机接口从一开始到最终的应用需要经过几个月的训练,因此,分类过程将随着训练数据的增多而降低时间效率。而且,目前越来越多脑机接口系统运行于普适环境,而普适环境下很多设备的计算性能并不高,更容易遇到实时响应瓶颈。因此,提高分类器的实时性能对于实时的应用系统至关重要。在本文中,我们采用MPI (Message Passing Interface)并行计算的框架,将分类算法实现并行化,以降低分类过程的时间消耗,从而达到实时应用的需要。
[Abstract]:Attention is a concept of psychology and a subject of research in brain science and bioinformatics. Monitoring people's attention can become an auxiliary tool for people to learn or work, thereby improving their learning and working efficiency, and even reducing the number of serious mistakes, such as learning in the process, If there is a tool to monitor students' attention, the teacher can remind the students when they are not focused, and the students can correct their mental state in time. In combination with computer science and biological information science, we can monitor human attention through brain-computer interfaces. It is a hot research topic to use computer to develop various kinds of BCI. However, at present, most of BCI's types are action types, that is to say, BCI is used to monitor what kind of action the brain is imagining. Then the Brain-Computer Interface (BCI) sends out the same action instructions to the external operation. The typical application of this BCI is the robot arm. However, there is a lack of research on the Brain-Computer Interface (BCI) for the mental state of monitoring attention in real time, because attention involves a lot of psychological knowledge and the research process is more complicated. The development of a real-time monitoring attention application system is less, more is offline processing of the brain computer interface, because the analysis and processing of brain waves is very complex, each processing step in the middle takes a certain amount of time. It is difficult to monitor in real time. In this paper, the current BCI technology and basic theory are deeply analyzed, and a real-time monitoring BCI is developed independently by absorbing the experience and theoretical knowledge of BCI development. And it is applied to monitor the attention of people in learning. The FIR algorithm based on Hanning window is used to extract six eigenvalues and the KNN (k-Nearest Neighbor) algorithm is used to classify the EEG signals. BCI needs a lot of training before it can be applied. Usually, a BCI needs several months' training from the beginning to the end. Therefore, the classification process will reduce the time efficiency with the increase of training data. Moreover, more and more BCI systems are running in pervasive environments, and many devices in pervasive environments have low computing performance, which makes it easier to meet the bottleneck of real-time response. Therefore, it is very important to improve the real-time performance of classifier for real-time application system. In this paper we use the framework of MPI (Message Passing Interface) parallel computing to parallelize the classification algorithm so as to reduce the time consumption of the classification process and achieve the need of real-time application.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP334.7

【相似文献】