基于偏最小二乘法的事件相关电位单次提取研究

SINGLE EXTRACTION OF EVENT-RELATED POTENTIAL BASED ON PARTIAL LEAST SQUARES

为满足脑-机接口特征提取实时性以及临床脑电检测高效性的要求,探讨事件相关电位的单试次高效提取技术尤为重要。将小波分析、经验模态分解、极限学习机以及偏最小二乘(PLS)应用于仿真和真实脑电信号,完成特征提取。结果显示:仿真实验中,不同信噪比下PLS提取性能稳定,P300潜伏期误差小于4 ms;真实脑电中,PLS少次迭代,特征提取更为精确,峰值误差0.551μV,峰值潜伏期偏移量27 ms,均小于小波、经验模态分解以及极限学习机多试次迭代结果(P<0.01)。结果表明偏最小二乘法在事件相关电位单试次提取中具有显著优势。

In order to meet the requirements of real-time feature extraction of brain-computer interface and high efficiency of clinical EEG detection,it is particularly important to explore the single-trial and high-efficiency extraction technology of event-related potentials.Wavelet analysis,empirical mode decomposition,extreme learning machine and partial least squares(PLS)were applied to simulate and real EEG signals to complete feature extraction.The results show that the PLS extraction performance is stable under different SNR,and the latency error of P300 is less than 4 ms.In real EEG,PLS features are extracted more accurately with less iteration.The peak error is 0.551μV,and the peak latency offset is 27 ms,which are smaller than the results of multiple iterations of wavelet,empirical mode decomposition and extreme learning machine(P<0.01).The results show that the partial least squares method has significant advantages in single-trial extraction of event-related potentials.