Research On Denoising And Extraction Algorithm Of Visual Evoked P300 Potential

Brain computer interface (BCI) is one of the most active research areas in recent few years and it has very great application value. BCI is a communication system to transmit information between brain and the external environment directly. Its characteristic is not dependent on the brain’s normal output pathway which consists of the peripheral nerves and muscle. BCI provides a method of communication for patients who lose the movement and language ability but with normal brain function to communicate with the outside world. They can express their thoughts or operate the devices directly through the brain without the need of language or action. BCI also provides a new communication model for healthy people, and can help workers in special circumstances, BCI has a wide range of application areas.Event related potential(ERP) is a commonly used EEG signal in BCI system, the main components of ERP include N100, P200, N200 and P300. Among them P300 is widely and maturately researched and has intimate relation with attention, memory and other high psychological activity. In this paper, we designed a stimulation interface based on P300 which was similar to the classical character spelling system and also carried out some experiments to collect the evoked potential. Because the EEG signal is very weak, the signal to noise ratio is extremely low and the disturbance noise is always existence in data acquisition process which would influence the analysis and the research of EEG, this paper focuses on the study of P300 potential denoising, feature extraction and classification recognition algorithm. The main work is as follows:(1) This paper improved the traditional character spelling stimulation interface and selected four subjects to carry out the P300 brain evoked potential acquisition experiment in the laboratory environment.(2) This paper proposed a combination denoising algorithm of Independent Component Analysis(ICA) and Empirical Mode Decomposition(EMD) to remove the common EOG interference in the pretreatment procedure of the objected EEG signal. ICA method used alone cannot automatically identify the artifact component and will cause the loss of some useful EEG information. The algorithm proposed solved these two problems, it had a better denoising effect and didn’t need to set the EOG reference signal in the experiment, which was easy to operate and had superior performance.(3) This paper combined wavelet decomposition and time domain energy entropy two algorithms to extract P300 feature of the de-noised and average EEG signal. Selected wavelet decomposition approximated coefficients and time domain energy entropy of the forth to sixth segment data as feature vector which can characterize P300 potential, reduced the data dimension and the amount of computation. Support vector machine (SVM) was chosen as the classifier for classification and identification of the feature vector. Results showed that the classification accuracy under two kinds of features was higher than single feature. It proved that the algorithm in this paper can effectively improve the recognition accuracy of the classifier.