Classification Method Of Motor Imagery EEG Signals Based On Deep Learning

As a new,multi-disciplinary technology,the brain-computer interface technology for motor imaging has important research significance.An important indicator to measure the performance of the brain-computer interface system for motor imagination is the accuracy of EEG signal recognition.Therefore,in order to improve the classification performance of the existing methods,we carried out a deep learning algorithm for the classification of motor imagery eeg signals based on the public data set in Physio Net.The following is the main research work of the paper:(1)In this paper,according to the frequency range of the motor imagery EEG,the original data set is filtered through a band-pass filter of 8Hz-30Hz;then ICA independent component analysis is used to remove artifacts;finally,the single experimental data is divided into time slices to achieve the effect of data reconstruction.(2)For the preprocessed data set,the Bi GRU motion imagination EEG classification model based on attention mechanism(ABi GRU)is studied.The model can extract more differentiated time-domain features from EEG signals through bidirectional GRU and attention layer.By compared with other methods,it is found that this model is indeed feasible.In general,the ABi GRU classification model proposed in this paper can alleviate the phenomenon of gradient disappearance and explosion to a certain extent owing to the special structure of GRU,and has achieved good classification results,but there are still shortcomings and need to be further perfected.(3)On the basis of work(2),a new classification model based on abigru densenet is proposed.The model combines abigru network with densenet network.Firstly,the time-domain features of EEG signals are extracted through abigru network,and then the feature matrix is used as the input of densenet network to further extract the spatial features in the signals.Finally,it is classified,and the average accuracy can reach86.73%.Its classification performance is better than most existing methods,and it also avoids the complex feature extraction process of machine learning algorithm,and has good individual adaptability and robustness,which provides a new reference for the classification of motor imagination EEG signals.