Research On Eeg Recognition Algorithm Of Motor Imagery Based On Deep Learning

With the rapid development of deep learning,researchers have gradually applied it to the field of brain-computer interface(BCI),and initially demonstrated its advantages over traditional machine learning.However,its application still faces many challenges,and the recognition rate of EEG is still a bottleneck that restricts the development of motor imagery brain-computer interfaces(MI-BCI).Therefore,this paper mainly focuses on the recognition algorithm of MI-EEG signal based on deep learning,and the main research works are as follows:(1)A DSC-ConvLSTM model based on attention mechanism is proposed for multi-classification of MI-EEG signals.Firstly,in the data preprocessing part,the sliding window is used for data enhancement,and it avoids the network to learn the difference features among sample trials;In terms of feature extraction and classification,deep separable convolution(DSC)is firstly used to extract spatial features of EEG signals.On the one hand,this reduces the number of network parameters and improves the system response speed.On the other hand,this network structure is more conducive to the direct extraction of EEG signal features.Secondly,the internal structure of LSTM unit is improved by using convolutional and attentional mechanisms,and the time domain features of EEG signal were extracted by using bidirectional convolutional long-short term memory network(ConvLSTM).ConvLSTM integrated the feature extraction capability of CNN and the temporal learning capability of LSTM.Finally,the attention mechanism is introduced to further improve the decoding performance of the model.The experimental results show that the average classification accuracy of the model in this paper reaches 73.6%and 92.6%on two datasets,BCI Competition Ⅳ datasets 2a and High Gamma Dataset,respectively,and it is superior to other algorithms in different datasets and different subject data,reflecting the robustness and effectiveness of the model.(2)A classification model transfer algorithm based on federated aggregation is proposed.By studying the influence of individual differences on EEG signal classification,a new theoretical framework and technical method are proposed.The overall framework adopts the deep transfer learning approach based on generative adversarial networks(GAN),and the generative adversarial transfer algorithm is designed from both unsupervised and semi-supervised cases.Federated aggregation is introduced to train the joint EEG classification model,which solves the problem of data leakage and data fusion during data heterogeneity.Finally,cross-subject experiments are carried out on the BCI Competition IV 2A dataset,and the results show that the proposed joint classification model greatly improves the cross-subject classification accuracy.It can be seen that the model proposed in this paper improves the influence of individual differences on the classification effect to a certain extent,and solves the problems of universality and portability of EEG classification model,which is of practical significance to promote the development of braincomputer interface technology to the direction of practicality and marketization.