高性能视觉ERP脑机接口若干问题研究

发布时间：2018-05-09 08:13

本文选题：ERP脑机接口 + 周期性干扰　；参考：《大连理工大学》2016年博士论文

【摘要】：最近十几年来,随着以神经信息学为核心的“人类脑计划”的开展,脑机接口技术在世界范围内引起广泛的重视。脑机接口依靠从人脑神经信号中获取的信息,实现人脑对外部设备的直接控制,创造了人与外部环境之间新的交流方式。视觉事件相关电位(Event-Related Potential, ERP)脑机接口是脑机接口研究的重要分支。具有高精度与高输出速率的高性能视觉ERP脑机接口一直是研究的重点。本文针对ERP脑机接口理论与应用研究中的性能改善问题、周期性干扰问题、异步控制问题,以及在线分类问题等重要问题展开研究,主要研究工作包括：(1)首先提出一种新颖的基于多变几何图形刺激的ERP脑机接口范式,然后,通过开展与经典范式的对比实验,研究多变几何图形刺激下脑机接口性能改善情况与ERP成分变化情况。实验与数据分析结果表明：所提出范式显著提高了ERP脑机接口的分类精度、信息传输速率与符号拼写率；在新范式下,前额-中央区的N2成分、P2a成分,以及枕-颞区的N1成分、P2b成分得到显著增强。进一步地,从应用角度对新范式下缩减电极数量的问题进行研究。数据分析结果表明,多变几何图形刺激可有效地用于缩减脑机接口分类识别所需的电极数量。(2)针对ERP脑机接口的周期性干扰问题,建立理论分析模型,认为这一周期性干扰主要是由靶刺激所诱发ERP响应的混叠效应引起的。而后,提出Toeplitz方法与差异波方法对周期性干扰进行抑制。实验数据分析结果表明,Toeplitz方法可以从非靶刺激平均信号中分离出靶刺激的ERP响应,而差异波方法则可以有效抑制周期性干扰。虽然周期性干扰对ERP成分的分析造成了不利影响,但是从另一个角度考虑,这一准周期波动却可以作为脑机接口处于控制状态的标志,用于解决脑机接口的异步控制问题。通过开展模拟的异步控制实验,表明了所提出基于准周期波动的异步控制方法的潜在价值。此外,对ERP脑机接口信号功率谱中出现的与周期性干扰相近的SOA (Stimulus Onset Asynchrony,刺激起始间隔)扰动线谱现象及其特性进行了理论模型与实验的研究。(3)将再生核希尔伯特空间内自适应投影次梯度方法(Adpative Projections Subgradient Method, APSM)应用于ERP脑机接口的在线分类问题,并提出改进的在线APSM-LM算法。使用实测数据分别对APSM-LM算法与基于高斯核函数的APSM-G算法中不同参数对在线分类性能的影响情况进行了研究。同时,从在线分类性能与处理时间角度,将APSM-LM算法与批处理的逐步线性判别分析方法进行对比。结果表明APSM-LM算法获得了接近于批处理算法的分类性能,而计算效率却显著提高。
[Abstract]:In recent years, with the development of the "Human brain Project" with neuroinformatics as the core, the brain-computer interface (BCI) technology has attracted wide attention in the world. The Brain-Computer Interface (BCI) relies on the information obtained from the human brain nerve signal to realize the direct control of the external equipment by the human brain and creates a new way of communication between the human being and the external environment. Visual event-related potentials (Event-Related potential) BCI is an important branch of BCI research. High performance visual ERP brain-computer interface with high precision and high output rate has been the focus of research. In this paper, the performance improvement problem, periodic interference problem, asynchronous control problem and on-line classification problem in the theory and application of ERP brain-computer interface are studied. The main research work includes: (1) A novel ERP brain-computer interface paradigm based on polygonal geometry stimulation is proposed. To study the improvement of BCI performance and the change of ERP composition under polygonal geometry stimulation. The experimental and data analysis results show that the proposed paradigm can significantly improve the classification accuracy, information transmission rate and symbol spelling rate of the ERP brain-computer interface, and, under the new paradigm, the N _ 2 component of the prefrontal to central region is composed of P2a. The N 1 component P 2 b of occipito-temporal region was significantly enhanced. Further, the problem of reducing the number of electrodes in the new paradigm is studied from the perspective of application. The results of data analysis show that the polygonal geometric stimulation can be used to reduce the number of electrodes needed for the classification and identification of brain-computer interfaces. (2) to solve the periodic interference problem of ERP brain-computer interface, a theoretical analysis model is established. It is considered that this periodic interference is mainly caused by the aliasing effect of ERP response induced by target stimulation. Then, Toeplitz method and differential wave method are proposed to suppress periodic interference. The experimental results show that the Toeplitz method can separate the ERP response of the target stimulus from the average signal of the non-target stimulus, while the differential wave method can effectively suppress the periodic interference. Although periodic interference has a negative effect on the analysis of ERP components, from another point of view, this quasi-periodic fluctuation can be used as a sign that the BCI is in a control state and can be used to solve the asynchronous control problem of BCI. The potential value of the proposed asynchronous control method based on quasi-periodic fluctuation is demonstrated by the simulated asynchronous control experiment. In addition, In this paper, the phenomena and characteristics of SOA Stimulus Onset Asynchronous, which is similar to periodic interference in the power spectrum of ERP brain-computer interface signal, are studied theoretically and experimentally. Adaptive projection subgradient method is applied to the online classification of ERP brain-computer interface. An improved online APSM-LM algorithm is proposed. The effect of different parameters in APSM-LM algorithm and APSM-G algorithm based on Gao Si kernel function on the online classification performance is studied by using the measured data. At the same time, from the point of view of online classification performance and processing time, the APSM-LM algorithm is compared with the stepwise linear discriminant analysis method of batch processing. The results show that the classification performance of the APSM-LM algorithm is similar to that of the batch algorithm, but the computational efficiency is improved significantly.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN911.7;R318

【相似文献】