【作者】 刘华生；

【导师】 汤井田；

【作者基本信息】 中南大学 ， 生物医学工程， 2011， 硕士

【摘要】 近年来,脑-机接口研究的兴趣持续强劲增长。脑-机接口系统使其用户可以通过大脑活动来控制外部设备,它把用户的意图翻译成设备的控制命令,为用户提供了一条新的控制或交流通道,该通道不依赖于外周神经系统和肌肉等大脑正常的输出通道。时-频分析能够把信号的能量分布描述为时间和频率的函数,是非平稳信号处理的有力工具。短时傅里叶变换、小波变换和wigner-ville分布是脑-机接口系统中三种常用的时频分析方法,论文从时间分辨率和频率分辨率角度讨论了这些方法的优缺点,并通过一些测试来深入讨论它们的优缺点及参数的选择。作为一种张量分解方法,平行因子分解保留数据的所有维,能够产生唯一的分解和具有物理意义的分量。论文把来自于不同通道脑电信号的时频分布组合成张量数据,其结构为通道×频率×时间,并通过拟合平行因子模型从这些张量数据中提取事件相关(去)同步化模式。为了评估这些特征的分类能力,采用了基于支持向量机、贝叶斯理论和隐马尔可夫模型的分类器并对它们进行对比。从结果可知,对于不同的时-频分析方法,短时傅里叶变换与小波变换的性能接近,而wigner-ville分布由于受干扰项的影响,性能稍差；参数的选择对这三种方法都很重要,尤其是短时傅里叶变换对窗函数的长度很敏感,而小波变换则能产生鲁棒的结果；对于分类器,改进的贝叶斯分类器通过整合不同时间处的信息而获得了比其它分类器更好的分类精度。更多还原

【Abstract】 Recently years, the interest in BCI research is strongly increasing. A Brain-Computer Interface is a system that allows its user to control external devices with brain activity, it translates user intents into device control commands, thus offering a new channel of control or communication that doesn’t does not depend on the brain’s normal output channels such as peripheral nerves and muscles.Time-frequency analysis is used to describe the distribution of signal energy as a function of both time and frequency, and is a powerful tool for non-stationary signal processing. short time Fourier transformation, wavelet transformation and Wigner-Ville distribution are three major time-frequency analysis approaches used in BCI systems, this article explained the merits and defects of those methods from the time resolution, frequency resolution, the advantages and shortcomings of these methods, and choosing of parameters are explained deeply by test cases.Parallel Factor Analysis(PARAFAC), a tensor decomposition approach, retains all dimensions and produces unique decomposition and physically meaningful components. Time-frequency distributions of EEG from different channels were grouped together to form data tensors of structure channel×frequency×time, and from which event-related (de)synchronization patterns were extracted by fitted to PARAFAC model. To evaluate the discriminative power of these features, classifiers designed by Support Vector Machine (SVM), Bayes theory and Hidden Markov Models (HMM) are introduced and compared with each other.For different time-frequency analysis methods, short time Fourier transformation and wavelet transformation revealed similar performance, and wigner-ville distribution analysis performed slightly worse than them because of interference terms. As indicated by results, selecting of parameters is important for all three methods, but short time Fourier transformation is sensitive to window length, and wavelet provides the most robust results. As classifier, through combining the information across time, modified Bayes theory based classifier made the highest accuracy.更多还原