Research On Classification Of Motor Imagination EEG Signal Based On Convolution Neural Network

Electroencephalogram(EEG)is the spontaneous electrical activity of the organism.It is a kind of physiological signal with time-space characteristics.The change of human emotion,the generation of different ideas and the making of various actions will cause the changes of EEG.Therefore,it has become an important research direction of brain computer interface(BCI)to explore the characteristics of EEG and build a bridge between patients with severe motor disability and the outside world.However,due to its weak amplitude and low signal-to-noise ratio,the research on EEG is facing many challenges.This paper summarizes the existing EEG feature extraction and classification methods.Aiming at the shortcomings of the existing methods,which can not take into account the time-space characteristics of EEG,but can only extract features from a single aspect,and the recognition rate is low,two classification methods based on convolutional neural networks(CNN)are proposed.The main work and innovation of this paper are summarized as follows.1.A new network architecture,Light Nets,is designed,and CNN structure is built on this architecture to realize the classification of single motion imagery EEG.First of all,a new lightweight network architecture named Light Nets is designed to optimize the direction of information flow and the way of information fusion in CNN computing.Secondly,according to the time and space characteristics of motion imagery EEG,a six layer CNN structure is designed based on the architecture,and the CNN is applied to the public data set and experimental data set to establish the classification model.Finally,the classification results are compared with the other three traditional methods.Experiments show that the method can effectively improve the classification accuracy.2.A new EEG classification method based on CNN and Bi Lstm(bidirectional long short term memory)is proposed.Firstly,according to the characteristics of CNN and Bi Lstm,which are more suitable for processing spatial and time series,the CNN-Bi Lstm structure is constructed.Secondly,the structure is applied to public data set and experimental data set to establish the classification model.CNN is used to extract spatial features,and Bi Lstm is used to extract temporal features.Add a classification layer to classify the extracted features.Finally,the classification results are compared with the other three methods.Experiments show that this method can improve the classification accuracy more effectively.