Motor Imagination EEG Signals Classification Based On Attention Mechanism And Deep Learning

Brain Computer Interface(BCI),as a new human-computer interaction technology,is an interdisciplinary research topic involving neuroscience,signal processing,pattern recognition and other disciplines.The BCI system based on motor imagination is regarded as one of the most promising brain-computer interface systems.In order to classify the mapping relationship between EEG features and motor imagery based on machine learning method,existing methods still cannot take into account the temporalspatial domain features of EEG signals and the classification accuracy is difficult to improve.In this paper,the feature extraction method of bidirectional long short term memory network based on attention mechanism and the convolutional neural network classification method of bidirectional long short term memory network based on attention mechanism are studied.The main work of this paper is summarized as follows:(1)The feature extraction method of bidirectional long short term time memory based on attentional mechanism is studied.In order to make the extracted EEG signal features better describe the temporal characteristics of the signals,this paper proposes a motor imagination EEG feature extraction and classification network(named ABiLSTM)based on the BILSTM model of attention mechanism.This model uses a combination of bidirectional long short term memory network and attention mechanism to extract more explanatory deep features.By comparing with four existing feature extraction methods,it is verified that the feature extraction method of BILSTM based on attention mechanism is effective.However,this method still has some shortcomings in classification performance.(2)A convolutional neural network algorithm for EEG classification based on the attention mechanism of bidirectional long short term memory is studied.In view of the characteristic that EEG signals have both temporal domain and spatial characteristics,in order to overcome the shortcomings of the above proposed method,this paper proposes and designs an EEG signal classification model(named ABiLSTM-CNN)that combines ABiLSTM and CNN neural network.Firstly,the temporal characteristics of EEG signals are extracted by ABiLSTM module.Secondly,the temporal-domain feature matrix is taken as the input of CNN module,and the temporal-spatial feature matrix is extracted by means of convolution operation and pooling dimension reduction.Based on the above research results,the ABiLSTM-CNN classification model constructed in this paper achieves an average accuracy of 90.72% on the Physio Net EEG motor imaging dataset.The results show that the ABiLSTM-CNN classification model of motor imagery EEG signal has a very good classification accuracy.