Research And Implementation On Classification Of Motor Imagery EEG Signals Based On Deep Learning

Brain Computer Interface(BCI)technology allows their users to control external devices using the conscious activity of the brain rather than the brain’s normal output pathways of peripheral nerves and muscles.BCI is a promising new direction in human computer interaction.Motor imagery electroencephalogram(MI EEG)based BCI system is a very important branch.Aiming at the practical application of MI EEG based BCI technology in intelligent wheelchair,this study studies feature extraction and pattern recognition of MI EEG using deep learning technology.In response to the shortcomings of current research in extracting single features,ignoring temporal information,and low classification accuracy,a multi-channel feature fusion method based on complex Morlet wavelet(CMW)and a MI EEG recognition method based on parallel CNN-Transformer are proposed.The main research work is as follows:In response to the problems of single extracted features and low classification accuracy of traditional methods,a multi-channel feature fusion method based on CMW is proposed,and a convolutional neural networks(CNN)model is constructed based on it.Firstly,the continuous wavelet transform(CWT)using the CMW is used to obtain the time-frequency map of MI EEG.Then,the time-frequency maps of Mu and Beta bands of multiple channels are cropped.Finally,the cropped time-frequency maps are stacked vertically to obtain a multi-channel fused feature map containing time,frequency and position information.The multi-channel feature maps input into CNN model are classified on public datasets,and the results are compared with traditional methods.The results show that the classification accuracy of the proposed method reaches 80.0%,which shows that this method can better represent the characteristics of MI EEG signals,and it can effectively improve the classification effect.To address the problem that CNN only utilize local spatial information and ignore temporal information when recognizing MI tasks,a MI EEG recognition method based on parallel CNN-Transformer is proposed.The method uses a CNN to extract spatially adjacent frequency and location information,and the Transformer to capture the internal connections of temporal information.The outputs of the two networks are spliced into fused feature vectors,and the Softmax function is used to classify the input vectors.Comparative experiments on public datasets show that the parallel network improves feature diversity and better classification performance compared to a single CNN or Transformer.Compared with other methods,the parallel network has higher classification accuracy and Kappa value of 84.6% and 0.685,respectively.The intelligent wheelchair control system based on BCI is built.The control scheme of using the subject’s intention to control the wheelchair movement is designed.The effectiveness and practicality of the MI EEG classification method proposed in this thesis are verified by testing and analyzing the online recognition rate and the travel trajectories of wheelchair.