Research On Feature Extraction Algorithms Of Electroencephalogram

Brain-Computer Interface(BCI) is a system which can realize the communication between brain and computer. BCIs do not depend on the brain's normal output channels of peripheral nerves and muscles. BCIs can help paralyzed patients control peripheral devices and communicate with the outside world. A series of neural electrical activities can be produced by Awareness or stimuli. It can be detected and identified by signal processing and pattern recognition. P300 is an evoked potential, and is stimulated by visual, auditory or pain. P300 potentials are different from different stimuli, which can be used to spell characters. The accuracy and velocity are vital important to characters spelling. BCIs have broad application prospects in biomedical engineering, military, business, criminal investigation and other fields.P300 speller paradigm is actually a multi-class recognition problem, mainly including preprocessing, feature extraction and recognition classification. Three aspects are all studied in this thesis. It employs low pass filter for signal preprocessing and SVM(support vector machine) for classification. ICA(independent component analysis) and wavelet are used to extract feature. Actually, feature selection is an optimization problem. It is significant to study feature vector. This thesis focuses on feature extraction and feature selection. Fisher discriminant criterion can be used for feature selection. The classification results show that Fisher discriminant criterion combination of feature extraction has a better performance than alone.