基于小波变换和盲源分离的P300识别算法研究

摘要

提出一种P300脑电信号识别的新方法,克服了多导联、多特征造成的操作繁琐、数据冗杂等问题.首先,针对P300小波变换中小波基如何优选缺乏必要的理论依据,提出结合电极分布特性,依据SNR和RMSE量化指标选择最优小波基,并结合SPWD时频分析对相干平均后的原始信号进行去噪处理,利用JADE盲源分离算法对得到的观测信号进行分离;其次,针对P300盲源分离后如何自动优选有效分量避免过度分解,提出结合G1法构建时空分析模型,最优化自动提取P300分量并映射回头皮电极处;最后,为改善BCI系统的在线应用,将EA和SFFS法相结合优选特征,构建具有6维特征向量的训练模型,利用C-SVM进行识别.实验结果表明,相对于传统的数据处理方法,P300成分的提取效果、系统分类精度和速度均有显著提高.%A new P300 EEG recognition algorithm is proposed.Complex operation and miscellaneous data of multichannel and multi-featured are avoided.First, aiming at selecting optimum wavelet base, which is lack of theoretical basis in wavelet transform of P300, a method is proposed based on SNR and RMSE, and the noise of original signals coherent averaged is removed combining the result of SPWD time and frequency analysis, and observations are decomposed by the preferred JADE algorithm.Then, aiming at selecting automatically and avoiding excess decomposition after BSS of P300, combining G1, temporal and spatial analysis model is built, P300 component is optimum extracted automatically and mapped to electrodes.Finally, in order to improve BCI system application online, combining EA with SFFS, and the training model of 6-dimension feature vector is built so that classified and recognized by C-SVM.As is shown by the experimental results, compared with traditional data processing technique, the effect of P300 component extraction, accuracy and speed of system recognition are improved visibly.