Research On Brain - Computer Interface Based On Motion Imagination

Brain-machine interface is to establish a separate pathways between the brain and the external environment through a computer or other electronic devices.It does not depend on muscle and peripheral nerve and directly implement communication and control.Brain-machine interface provides a new way to communicate with the outside world for the loss of some or all of the patients muscle control function.This paper mainly studied the eeg data processing methods.On the use of the eeg signals processing preprocessing, feature extraction and pattern classification algorithms in detail.At the same time, the movement of two classification by off-line imagine eeg data sets to verify this algorithm.At first the thesis introduces the concept of brain-machine interface and system composition.The rise of the brain-machine interface research field and the current research situation and application results are all summarized.The thesis elaborated the brain electrical signal preprocessing, feature extraction and pattern classification algorithms.The effectiveness of the proposed feature extraction algorithm directly affects the classification result.Wavelet decomposition and wavelet reconstruction technology are used to multi-scale analysis of motion imagine eeg signals.The most obvious signal features in each spectrum is extracted.The signal is processed by average.Combining cumulative wavelet entropy and AAR algorithm, to extract more characteristics of eeg signals, improved the subsequent classification accuracy.The results show that,Analyzing the sub-band alone can extract hidden in the movement to imagine more accurate information in eeg signals.Combined with the AAR model parameters, the classification accuracy has improved significantly.Selection of classifier is particularly important.Neural network based unscented kalman filter(UKF) is used for a classifier to identify both brain electrical characteristics.UKF algorithm through rapid iterative computation can accurately estimate the weights of neural network, and complete the training of neural network.Results show that the classifier’s classification performance is superior to the commonly used linear discriminant analysis (LDA) and SVM classifier.In allusion to two classification of eeg data for movement imagine,adopte the combination of a variety of feature extraction and classification algorithm and analysis the features of wavelet entropy under different rhythm. Combined cumulative frequency band energy feature with the AAR model parameters,lower classification error are obtained.In order to obtain higher classification accuracy,This paper applies neural network based on UKF in the problem of eeg data processing classifier,and compared with the classification results of LDA and support vector machine (SVM).Finally this paper introduces the BCI2000 platform,simulate BCI experiment based on sensorimotor rhythm,to simulate control by thinking.