【作者】 官金安；

【导师】 林家瑞； 陈亚光；

【作者基本信息】 华中科技大学 ， 生物医学工程， 2005， 博士

【摘要】 直接用大脑思维活动的信号与外界进行通信,实现“心与心”的交流,甚至达到对周围环境的控制,是人类自古以来就追求的梦想。脑-机接口(Brain-Computer Interface: BCI)这种新颖的人机交互模式提供了实现这一梦想的科学途径。人们希望这种全新的通信技术能够用于辅助控制交通工具、武器和其它系统,特别为那些神经肌肉受损,不能使用常规通信手段的残疾患者提供与外界进行交流的另一途径。所谓脑-机接口,是一个不依靠外周神经和肌肉组织等通常的大脑输出通道的通信系统。近5 年来,这一领域的研究逐渐形成了热点,世界上数十个研究小组已开发出多种形式的BCI 实验系统。其中已有3 种脑控键盘的报道。但它们的共同问题一是以低频闪烁方式提供视觉诱发信息,容易造成使用者的疲劳; 二是通信速度太低,只有5～27 比特/分钟,很难满足实际需要。有鉴如此,我们在国家自然科学基金的资助下,开展了基于“模拟自然阅读”诱发模式的脑控拼写装置的研究。试图使该系统的通信速率达到90 比特/分钟,并以更自然的方式给用户提供一种舒适的使用环境。在这个系统中,通信载体、信源编码、虚拟键盘的设计和脑-机接口信号的单次提取是四个最核心的问题。在前两个问题已基本解决的情况下,本文就后两个问题展开了深入的研究。研究内容、结果与创新点如下: 1. 调研了脑-机接口的起源、意义、定义、分类、信号特点、信号处理及模式识别方法、目前研究的现状及面临的挑战等。进行综合分析后,指出了目前存在的不足及其发展方向,提出了我们的解决方案。成果发表在附录1 的【1】、【2】、【6】中。2. 提出了一个新颖的“双页虚拟键盘”方案,并对其按键位置的排布进行了合理的设计。它弥补了常规脑控拼写装置中信源数量增多将导致选择单个信源时程偏长的不足。分析表明,这种设计在原有设计指标的基础上,通信速率将有约70%的提高,可达150 比特/分钟以上,是现有BCI 系统的5～20 倍。这部分的研究成果发表在附录1 的【4】、【5】中。3. 对BCI 通信载体信号进行了谱分析,发现非靶刺激与靶刺激所诱发的VEP 信号的相对功率谱,在5Hz 以下有较大的变化,可达15db 以上; 而在10Hz 以上却基本没有变化。这为后续降低特征维数,提高信号处理速度提供了理论更多还原

【Abstract】 It’s an ancient dream of human that using their mind to control or communicate with peripheral circumstance directly. Brain-Computer Interface (BCI) technologies revealed the scientific approaches to make the dream coming true. This novel method gives a valuable new option for individuals who cannot use conventional communication systems that depend on peripheral muscles and nerves, particularly those with neuromuscular disorders and motor disabilities. Brain computer interfaces give their users communication and control channels that do not depend on the brain’s normal output channels of peripheral nerves and muscles. In last five years, BCI is becoming a hotspot and have arisen great interesting of scientists all over the world. There several BCI systems came out, among them three mental controlled keyboard had been reported. These innovatory systems achieved an average speed rate about 5~27bits/min. But there are several aspects to be improved. Firstly, the speed rate is not so high; and secondly, the flash in a low frequency may cause eye fatigue rapidly. To amend these deficient, we investigated an INR SPELLER system based on a so-called Imitating-Natural Reading (INR) paradigm. It was demonstrated that the Exogenous Given Reactions which reduce the signal-to-noise ratio were restrained and spontaneous endogenous potentials were “regularized”significantly with this novel modality. The system is expected to have a speed of 90bits/min, and to give their users comfortable conditions in using it. There are four key issues in the system, that is communication carriers, source coding, designation of virtual keyboard, and the single-trial estimation of its message carriers. The present dissertation dedicated to have a thoroughly investigation to the latter two issues. The contexts, results, and innovations of this work are as follows: 1. By investigating the BCI’s origination, significance, definition, classification, feature of signals, signal processing, pattern recognition, and the development and challenges, etc., the future developments were clarified. 2. The communication carrier and coding methods were introduced. A dual-page virtual keyboard was proposed based on former works. The analytical results suggested that there are 70% higher in communication rate than the former designation target, it could up to a speed of 150bits/min, and will have a speed 5~20 times higher than the current BCIs. 3. The analysis of power spectrum of carriers in EEG showed that, the relative power have a significant change below 5Hz which could up to 15db; whereas, there were no change in the range of higher than 10Hz. The analysis gave theoretical guidelines for the following feature dimensional reducing and boosting the speed of communication. The features embedded in EEG were enhanced by means of AR model and wavelet filtering. This could improve the classification accuracy further more. we did another experiment to extract N2 components using independent component analysis (ICA). Facing the challenge of uncertainty of the polarity in ICs, a new algorithm with N2 bipolar threshold was proposed, and solved the problem. The N2 components were enhanced and made the single-trial estimation of N2 be feasible. 4. The pattern classification algorithm of support vector machine (SVM), which based on statistical learning theory, for the single-trial estimation of carriers was researched. We proposed an new SVM algorithm with AR model and ICA feature extraction. Applying the program written in Matlab6 to the data from three subjects, the effects of single-trial estimation of carriers were investigated thoroughly by means of various montages of multi channels, single-channel, difference time lengths, difference time intervals, etc. finally, a perfect results were gained by combination of P2, N2, and P3 components from two channels. The results also suggested the experiment paradigm of our mental speller is feasible.更多还原