Alternate Modes of Human-Computer Interaction: EEG Recognition with Neural Networks

今年是三年持续努力的第一年。本研究旨在探讨利用脑电(EEG)模式的在线识别来实现人机交互的可行性。一个可以在线识别一个人正在执行几项精神任务中的哪一项的系统可以提供一个字母表，严重残疾的人可以用它来编写对轮椅等设备的命令。之前的大多数研究都集中在识别的准确性上，而不是快速、实时的反应上。本研究的目的是分析现有脑电信号识别方法的局限性，利用神经网络方法提高脑电信号识别的准确率，减少计算时间，并开发一种实用的实时识别方法。将比较传统方法和针对高维数据设计的神经网络学习方法的性能。对特征和分类器特性的评估将基于分类精度、稳健性和计算速度。该项目总体上与模式识别领域有关，特别是与神经网络和人机交互领域有关。

This is the first year of a three year continuing effort. This research is to study the feasibility of human-computer interaction through the use of on-line recognition of electroencephalogram (EEG) patterns. A system that can identify, on-line, which of several mental tasks a person is performing could provide an alphabet with which a severely disabled person can compose commands to devices like a wheel chair. Most prior research has focused on recognition accuracy rather than on fast, real-time responses. The goals of this research are to characterize the limitations of current EEG recognition methods, to increase the accuracy and decrease the computation time of EEG recognition by use of a neural network approach, and to develop a practical real-time method. The performance of conventional methods and of neural network learning methods designed for high- dimensional data will be compared. Evaluations of feature and classifier characteristics will be based on classification accuracy, robustness and computation speed. This project has relevance to the field of pattern recognition in general, and to the neural network and human-computer interaction fields in particular.