项目名称: 多维脑信号分析方法及其在脑-机接口中的应用研究
项目编号: No.61305028
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 张宇
作者单位: 华东理工大学
项目金额: 25万元
中文摘要: 本项目针对基于脑电信号(Electroencephalogram, EEG)的脑-机接口(Brain-Computer Interface, BCI)在人-机交互系统中的理论和应用研究需要,以深入研究多维脑信号分析方法提高脑电模式分类效果为切入点,开发具有高鲁棒性和实用性的实时BCI控制系统。根据特定脑电模式,挖掘多维特征分析和降维的本质,对一维优化算法进行有针对性的多维融合,改善脑电模式特征分析效果,并通过多维特征降维和公用分类模型构建,减少BCI系统校验时间,提高实用性;利用贝叶斯推论和高斯过程的理论知识,研究多维脑信号分析的概率生成模型及核方法,增强其对BCI的应用能力,提高系统整体性能;根据EEG的多维神经生理学特性,设计合适的多维正则化处理方法,并基于集成学习和贝叶斯证据框架的理论知识,提出适用于小样本的有效多维超参数优化方法,增强多维脑信号分析算法的鲁棒性。
中文关键词: 脑-机接口;脑电信号;多维特征分析;多维正则化;模式分类
英文摘要: This project aims to develop a real-time brain-computer interface (BCI) system with high robustness and practicability, through in-depth research on multiway electroencephalogram (EEG) analysis methods to improve classification results of EEG patterns, according to the requirements of theory and application research for BCI in the human-machine interactive system. Essences of multiway feature analysis and dimensionality reduction are explored. Different one-way optimization algorithms are integrated into a novel multiway analysis method for the specific EEG patterns to improve results of EEG feature analysis. Multiway dimensionality reduction and generic classification model are exploited to reduce calibration time and enhance practicability of the BCI system. Probabilistic generative models and kernel methods for multiway EEG analysis will be developed based on the Bayesian inference and Gaussian process to enhance the capability of application to BCI, and hence improve the overall system performance. To improve the robustness of multiway EEG analysis methods, multiway regularization will be explored according to the multiway neurophysiological characteristics of EEG. Effective algorithms will be developed based on ensemble learning and Bayesian evidence framework for multiway hyperparameter optimization, espec
英文关键词: Brain-computer interface;Electroencephalogram;Multiway feature analysis;Multiway regularization;Pattern recognition