摘要： 开发基于脑－计算机接口的脑控拼写装置，核心问题之一是实时、准确地从头皮电极记录到的脑电背景信号中提取通信载体信号，以决定用户选择的是哪个按键。该文对利用独立分量分析从多通道脑电信号中提取N2 成分进行了研究，针对独立分量极性及通道不确定的问题，结合基于负熵非高斯性极大判据的ICA 算法，采用双极性阈值方法，较好地增强及提取了诱发脑电中的N2 成分，对比分析表明，该方法有效地提高了后续模式分类的正确率。

关键词: 独立分量分析；视觉诱发电位；脑-机接口；N2 成分

Abstract: A mental speller based on brain–computer interface is constructed for those with neuromuscular disorders and motor disabilities. One of the key issues is to precisely recognize the visual evoked potentials from ongoing Electroencephalogram (EEG) background that are recorded from scalp electrodes, and this is utilized to determine which key is “pressed” by user. The extraction of N2 from multi-channel electroencephalogram using independent component analysis (ICA) is investigated. The ICA algorithm based on maximizing nongaussianity measured by negentropy is employed to do the work. A bipolar-valve method is proposed to solve the problem of polarity and channel uncertainty in independent components. N2 components are enhanced and extracted efficiently. The results of subsequently comparison research show that ICA method can boost up the classification accuracy significantly.

Key words: Independent component analysis; Visual evoked potentials; Brain-computer interface; N2 components