用多通道特征组合和SVM单次提取诱发脑电信号

Single-trial estimation of visual-evoked potentials using SVM with multi-channel feature montages

以“模拟人类自然阅读诱发模式”产生的诱发脑电信号作为载体,利用脑-机接口这种新颖的人-机交互方式构建一种脑控拼写装置.在这种实时通信模式中,不能采用认知科学实验及临床中的常规相干平均方法来提取诱发电位,而必须实现特征信号的单次识别.对来自四个通道的各种信号成分进行特征组合,利用支持向量机分类器对一名被试者脑电信号中的载波成分进行了单次提取,最佳特征组合的平均正确识别率为98.8%,证明了诱发模式的先进性和系统实现的可行性.

Exploiting potentials induced by ＂imitating-human-natural-reading inducing paradigm＂ as communication carriers, a mental speller based on brain-computer interface was constructed. In this online paradigm, single-trial feature estimation should be used instead of grand average that usually used in the cognitive or clinical experiments. With several montages of component features from 4 channels, and with carefully parameters optimizing, the single-trial estimation of evoked potentials using support vector machines in a subject was explored, and gained perfect classification rate of 98.8 %. The results show the advantages of the inducing paradigm used in our mental speller.