脑机接口（Brain-computer Interface, BCI）是一种基于非生理传递和控制的技术，用于在体外建立人工通道，从而来完成大脑指令并执行人机交互的过程。正常的自然情况下，大脑指令的实现，是大脑信号产生后通过生理传输通道（神经系统）传输信号和生理应用端（肌肉组织等）来实现。而在脑机接口技术中，人工直接从大脑皮层采集脑电信号，通过解析信号转化为机器识别语言，再将其传输到机器终端实现大脑指令。随着信息产业的不断完善和互联网技术的不断发展，脑机接口技术越来越受到各国学者的关注与研究，认为会成为未来的朝阳产业。

       根据目前研究来看，主要的一个研究方向就是基于P300电位检测技术的研究和应用，并且已经取得很大的成果。但是，这种技术下的脑机接口系统的发展依然受制于两个主要影响因素：一个是信息传输率比较低，另一个就是准确率不太理想。因此，本文在基于P300的技术下，对这两个方面进行设计方面的优化，并且实现该系统并验证。

        首先针对信息传输率低的问题，本文提出了基于减少刺激序列出现次数的一种动态的停止准则的优化算法。其原理是选择支持向量机（Support Vector Machine，SVM）作为分类器分类，然后对每次刺激序列结束后得到的信号进行分类，将所得的每次返回值所占总的返回值的比例作为一个判断元素，将该元素与设定的初始阈值比较，超过阈值就停止本次目标字符的刺激序列的出现。如果没有达到阈值，就继续出现，直到原本设定的最大的次数，然后输出结果。该算法在实际实验中表现良好，在保证相对高的准确率的同时，刺激序列的出现明显降低，大大缩短了信息传输的时间，提高了字符的输入速度。

        其次，本文设计了一种新的刺激范式，该范式有效的提升了准确率。根据已知的研究表明，目标字符的出现概率决定了P300信号被检测到的概率，两者是负相关的关系，目前主要的刺激范式都是一种基于传统的RC经典范式（行/列交替出现的方式）。基于该原理，设计了一种基于区域的单字符闪烁刺激范式，这种范式中的目标字符出现概率更低，并且基于区域的范式，会减小旁边字符对目标字符的影响。

       最后，在传统的字符输入系统，大部分采取的是同步的手段，在一定的时间内进行字符的输入，而异步是由用户自主决定输入时间，相对来说更加的灵活。本文在异步上的算法是基于一种注意力水平的算法，设定一个注意力水平的阈值。当超过阈值时，系统就判断进入字符输入准备，开始接受用户的脑电信号的输入和分类，当小于阈值时，就判断该用户未进行字符输入活动，系统不再进行字符输入任务。本论文结合设计的优化的算法，成功的实现了整个系统的开发。

      Brain-Computer Interface (BCI) is a technology based on non-physiological transfer and control，for the establishment of artificial channels in vitro，to complete the brain instructions and the implementation of human-computer interaction process. In normal natural circumstances, the realization of brain instructions is realized through the physiological transmission channel (nervous system) and physiological application (muscle tissue, etc.) after the generation of brain signals. In the brain-computer interface technology, electroencephalogram signals are collected directly from the cerebral cortex and converted into machine recognition language by analyzing the signals, and then transmitted to the machine terminal to realize the brain instructions. With the continuous improvement of the information industry and the continuous development of Internet technology, brain-computer interface technology has attracted more and more attention and research from scholars all over the world, and it is believed that it will become a sunrise industry in the future.

        According to the current research, one of the main research directions is the research and application of P300 potential detection technology, and great achievements have been made. However, the development of brain-computer interface systems based on this technology is still subject to two major factors: the low information transmission rate and the unsatisfactory accuracy rate. Therefore, this paper optimizes the design of these two aspects based on P300 technology, and implements and verifies the system.

       Firstly, to solve the problem of low information transmission rate, this paper proposes a dynamic stopping criterion based on reducing the frequency of stimulus sequences. The principle is to select Support Vector Machine (SVM) as the classifier for classification, and then to each stimulation signal obtained by the end of the sequence classification, will use the proceeds of each return value accounts for the proportion of the one as an element, the element and setting of the initial threshold comparison, more than threshold, stop the stimulation of the target character sequence. If the threshold value is not reached, it will continue to appear until the maximum number of times originally set, and then output the result. This algorithm performs well in the actual experiment. While ensuring relatively high accuracy, the appearance of stimulus sequence is obviously reduced, which greatly shortens the time of information transmission and improves the input speed of characters.

       Secondly, this paper designs a new stimulus paradigm, which effectively improves the accuracy. According to the known studies, the occurrence probability of the target character determines the probability of detection of P300 signal, and the two are negatively correlated. At present, the main stimulus paradigms are all based on the traditional RC classical paradigm (row/column alternation).Based on this principle, a single-character scintillation stimulus paradigm based on region is designed, in which the occurrence probability of target characters is lower, and the region-based paradigm will reduce the influence of adjacent characters on target characters.

       Finally, in the traditional character input system, most of the means is synchronous, in a certain time for the character input, and asynchronous is determined by the user input time, relatively more flexible. In this paper, the asynchronous algorithm is based on an attention level algorithm, set a threshold of attention level. When the threshold value is exceeded, the system judges that the input of characters is ready, and begins to accept the input and classification of the user's EEG signals. When the threshold value is less than, it judges that the user has not carried out character input activities, and the system no longer carries out character input tasks. This paper combined with the design of the optimization algorithm, the successful realization of the whole system development.

目 录

摘 要 I
ABSTRACT III
第一章 绪论 1
1.1课题研究的背景和意义 1
1.2国内外研究现状 3
1.3本文主要框架 6
第二章 基于P300的BCI系统基础理论7
2.1 BCI系统的分类 7
2.2 基于P300的BCI系统 9
2.2.1 事件相关电位 9
2.2.1 P300电位介绍 9
2.2.2 基于P300的脑机接口系统 10
2.3 P300的信号采集 12
2.4信号处理技术 13
2.4.1 伪迹的处理 14
2.4.2 干扰信号的去除 15
2.5 特征提取 15
2.6 分类器的选择 17
第三章 异步动态停止算法设计 24
3.1 异步算法 24
3.1.1 EEG中信号频率 25
3.1.2 异步算法设计 26
3.2 动态停止算法 27
3.2.1动态停止算法的原理 28
3.2.2 阈值的选择 31
3.3 刺激范式的优化 33
3.4 小结 34
第四章 实验结果与分析 35
4.1 竞赛数据验证 35
4.1.1实验数据来源 35
4.1.2实验过程 36
4.1.3 实验结果 36
4.2 离线测试 40
4.2.1 离线实验方案 40
4.2.2 离线实验与结果 42
4.3 在线测试 46
4.3.1 在线实验过程 46
4.3.2 在线实验结果 48
4.4 小结 50
第五章 总结与展望 51
5.1总结 51
5.2展望 52
参考文献 53
致 谢 57
攻读研究生期间取得的成果 58