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

In the human body structure,the human brain has always been in the first place.It not only controls various behaviors in human daily life,but also endows human beings with various senses to contact the outside world,such as touch,vision,hearing,and so on.But in real life,some people have been suffering from the inconvenience of movement caused by brain injury.In order to help this group of people realize their desire to communicate with the outside world again,researchers put forward the concept of Brain-Computer Interface(BCI),which uses the relevant technologies of brain science and computer science to realize the intelligent interaction between patients and the outside world.The brain-computer interface technology based on Motor Imagery(MI)can analyze and process the patient’s electroencephalogram(EEG)signal,get the patient’s action intention and convert it into corresponding instructions,and control the external action equipment.However,the classification and recognition accuracy of the current motion imagination BCI system is somewhat deficient.How to improve the accuracy of BCI system classification task has become a hot topic in BCI field in recent years.In this paper,the internal principle of classification and recognition of motion imagination BCI system is deeply studied.It proposes effective algorithm models for feature extraction stage and feature classification and recognition stage,and designs and implements classification and recognition simulation experiments based on MI-EEG signals to verify the effectiveness of the proposed algorithm.Simulation experiments show that the two optimization schemes proposed in this paper have excellent performance in the feature extraction and feature classification recognition stages of the Motion Imagination BCI classification and recognition system.The main research work and innovation points of this article are as follows:1.Understand the physiological knowledge and related technologies involved in the classification and recognition system of the BCI system for motor imagery.It considers the basic composition of the BCI system from two aspects: the structure of the BCI system and the classification of EEG signal experimental paradigms,and introduces the physiological knowledge of EEG signals.At the same time,it analyzes the key event phenomena of motion imagination EEG signals,and analyzes the related technologies of motion imagination BCI classification and recognition system in three aspects:signal acquisition,signal feature extraction,and feature classification and recognition.2.Aiming at the classification system of motor imagery EEG signals based on deep learning,a feature extraction algorithm for EEG signals based on EEGNet neural network is proposed from the perspective of feature extraction of motor imagery signals in BCI systems.Based on the proposed signal feature extraction scheme,it considers the applicability of the deep convolution kernel of the EEGNet neural network to the field of motor imagery EEG signals,and analyzes the impact of two convolutions on the classification accuracy of classification and recognition systems.At the same time,traditional linear discriminant analysis(LDA)is used to reduce signal dimensions and KNearest Neighbors(KNN)is used for classification and recognition simulation experiments.Simulation experiments show that compared to the commonly used filter bank common spatial pattern(FBCSP)feature extraction algorithm,the classification accuracy of the motion imagination brain computer interface system based on the EEGNet neural network signal feature extraction algorithm has significantly improved,and the number of network parameters has been reduced.3.Aiming at the classification system of motor imagery EEG signals based on deep learning,a classification and recognition algorithm for EEG signals based on Auto Encoder(AE)and Transformer models is proposed from the perspective of classification and recognition of motor imagery signals in BCI systems.Based on the use of EEGNet neural network for signal feature extraction,the algorithm uses AE self encoders to reduce data dimensions,and finally uses the position encoding of the Transformer model to consider the location information of global features and uses the multi header self attention mechanism to consider the internal correlation of the feature matrix,with the goal of achieving high classification recognition rates.Simulation experiments show that the classification accuracy obtained by the AE-Transformer system is not only significantly better than the LDA-KNN algorithm,but also the algorithm complexity is better than the novel BCI classification system that combines convolutional neural networks(CNN)and Transformer.