Analysis Of Motor Imagery EEG Signals Based On Common Spatial Pattern And BP Neural Network

Brain-computer interface(BCI)is a communication system that is established between the human brain and computers or external devices without relying on the regular brain peripheral nerve and muscle systems.Electroencephalography(EEG)signals contain rich information about brain activity.By acquiring and analyzing EEG signals,extracting features and correctly identifying them,the brain can effectively control external devices.Motor imagery based BCI(MI-BCI)is one of the main directions of BCI research in recent years.Aiming at the problems of low classification accuracy and poor individual adaptability of MI-BCI,this paper proposed a MI-EEG analysis model based on common spatial pattern(CSP)and backpropagation(BP)neural network.The model can extract the feature and classify the EEG signals.The main research work is shown as follows:1.The development status of MI-BCI as well as the MI-EEG analysis algorithms in recent years have been introduced comprehensively.And the paper analyzes the event-related desynchronization/synchronization phenomenon that appears in EEG signal when doing the motor imagery.2.A model based on CSP and BP neural network was proposed for MI-EEG processing.Firstly,CSP is used to spatially filter the EEG signals,and then the temperal weight vector is introduced to extract the features of the filtered EEG signals.Using artificial neural network as the classifier,cross entropy is chosen as the loss function of neural network.The model adopts the backpropagation algorithm and the gradient descent method.The model automatically adjusts the temperal weight vector and the free parameters in the neural network in trainings,and continuously reduces the classification errors of the model,so that the model gets the best classification effect.3.The proposed model was applied for the 2-class and 4-class MI-EEG data sets in the BCI competitions,and the average classification accuracy obtained are 90.0%,82.1%and 71.6%respectively.The results are better than the competition champion and convolutional neural network algorithm.For the individual samples that obtained poor classification results in the competition,the classification accuracy obtained by this model is increased by 2%?11%compared with the competition champion.The experimental results show that the proposed model improves the classification accuracy of 2-class and 4-class MI-EEG,and the model shows good adaptability among different individuals.The validity of the proposed model for the classification of MI-EEG is verified.