Classification Of Motor Imagination EEG Signals Based On Convolutional Neural Networks

In recent years,brain computer interface technology has become a new way of human-machine interaction.Due to the different ways of collecting EEG signals,brain computer interface systems have developed many different research directions.Among them,research on the classification of motor imagery EEG signals has a great promoting effect on promoting the application and development of brain computer interface technology.However,there are still the following issues in current research on the classification of motor imagery EEG signals: 1.The signal-to-noise ratio of noninvasive collected EEG signals is low.2.Traditional temporal signals of motion imagination cannot effectively reflect the characteristics of motion imagination.3.The differences between individual subjects also have a significant impact on the classification of motor imagery EEG signals.The main work of this article is to address the above issues as follows:(1)To address the issue of noise interference in motor imagery EEG signal data,this paper first removes the data from the EEG channel,filters out useless signals through FIR filters,filters out EEG artifacts using ICA independent component analysis,and then overlays and averages the EEG signal to make it smoother.This improves the signal-to-noise ratio of EEG signals.(2)In response to the problem that traditional EEG signals cannot effectively reflect motion imagery,the time-frequency images of motion imagery EEG signals contain rich frequency and energy feature information.Therefore,this article proposes the use of short-time Fourier transform and continuous wavelet transform timefrequency analysis methods to transform motor imagery EEG signals into a series of topologically preserved time-frequency maps,and then uses a combination of convolutional neural networks and short-term memory networks to extract the features of the time-frequency maps.This method was evaluated in BCI 2b two classification tasks,and the accuracy was improved by 1% compared to the CNN ASE method,indicating that this method can effectively extract features from time-frequency maps.(3)To address the issue of significant differences between individuals,in order to fully extract the temporal and spatial features of EEG signals,this paper proposes a combination of wavelet packet decomposition(WPD)and common spatial pattern(CSP)feature extraction methods,which jointly extract the time-frequency and spatial features of motor imagery EEG signals,overcoming the shortcomings of traditional EEG signal feature extraction methods.This method achieved an accuracy of 71% and a kappa coefficient of 0.61 on the BCI 2a dataset,indicating that it can effectively extract the time-frequency and spatial features of the data.