Research On Signal Processing Algorithm Of P300 Brain-Computer Interface Based On Blind Source Separation

Brain-computer interface(BCI) is a communication system, which is built between a human brain and a computer or other electronic devices without depending on conventional information output paths of the brain. P300 electroencephalography(EEG) is a particular endogenous component of event-related potential, and widely applied in BCI systems. The method to extract and recognize P300 component accurately and quickly, which is a hot issue in the research of BCI, is of great significance to cognize bioscience, neuroscience, teleinformatics and other scientific fields. In this paper, in terms of pretreatment, feature extraction, feature optimization and classified recognition, blind source separation(BSS) that is widely used in the field of signal processing in recent years is used to analyze P300-based BCI signal processing algorithm.Aiming at the process of pretreatment on P300, the spectrum of artifact components overlaped with P300, the optimization of wavelet base lacked of theoretical basis in wavelet transform, and source signals decomposed in excess, estimation signals selected artificially after BSS, a new P300 EEG extraction algorithm is proposed. Firstly, combining distribution characteristic of electrodes, optimal wavelet base is selected based on quantitative indicators of signal to noise ratio and root mean square error. Secondly, according to the result of smoothed pseudo Wigner-Ville distribution, original signals are filtered in time, frequency and spatial domain by coherent average, wavelet transform, and BSS algorithm. Lastly, based on inverse of separation matrix, global and local pearson correlation coefficient, local average amplitude and local peak amplitude, temporal and spatial analysis model is built combining G1, P300 components are optimum extracted automatically and mapped to eletrodes. The performance of four typical BSS algorithms, that includes Fast ICA, Informax, SOBI and JADE, is compared in the process of P300 potential extraction. As is shown by the experimental results, compared with traditional signal processing technique, the effect of P300 potential extraction with the proposed method is improved visibly.Aiming at the process of feature extraction, feature optimization and classified recognition on P300, complex operation and miscellaneous data caused by multichannel and multi-featured, a new P300 EEG recognition algorithm is proposed. First of all, combining enumerative with sequential floating forward selection and according to separability criterion based on scatter matrix, initial features that is generated in the process of wavelet transform and BSS processing are selected. Next, nonlinear soft margin support vector machine is utilized, and parameters of classifier are optimized by cross validation and particle swarm optimization, then the training model of 6-dimension feature vector is built so that classified and recognized. As is shown by the experimental results, accuracy and speed of system with the proposed method is greatly improved, and the foundation of P300-based BCI system application online is laid.