Research On EEG Signal Classification Method Based On Convolutional Neural Network

Electroencephalogram(EEG)signal classification is the key in the research of emotion recognition,brain-computer interface system,fatigue driving,and various brain disease detection.The accuracy of classification directly affects its performance.The EEG signal has the characteristics of low frequency and weak amplitude,which makes the collection of EEG signals extremely susceptible to interference,resulting in more complex EEG signal components,and increasing the difficulty of EEG signal classification.Now researching new EEG signal feature extraction methods to build a more robust EEG signal classification model has become a hot spot in current research.Aiming at the problems of low classification accuracy,complicated and time-consuming methods of the existing convolutional neural network EEG signal classification model.Two sizes of convolution kernels are added to the layer,and a multi-scale convolution kernel convolution neural network EEG signal classification model is proposed.The multi-scale convolution kernel convolutional neural network can extract more comprehensive base-level features of EEG data from different dimensions,which greatly improves the performance of the EEG classification model.The main contents include the following:(1)The plateau EEG data set used in this paper was preprocessed with potential re-reference,artifact removal,and filtering to obtain EEG data with high signal-to-noise and meeting network input requirements.(2)A multi-scale convolution kernel convolution neural network EEG classification model is proposed,and the classification experiment of plateau EEG signals is completed.The validity of the proposed model was verified in the EEG signal two-class and multi-class experiments.The experimental results show that the accuracy rate of the second classification can reach 92.09%,the accuracy rate of the third classification can reach 89.96%,and the accuracy rate of the fourth classification can reach 80.15%;In addition,in the experiment of two-to-two classification of EEG signals,it was found that there are differences in EEG signals at different altitudes,and there is a positive correlation with the difference in altitude.There is a threshold for the influence of EEG signals.(3)In order to improve the performance of the EEG classification model,a three-layer stacked sparse self-encoding network was designed for feature extraction of EEG signals,which was verified by classification experiments on handwritten data sets and EEG data sets.The multi-scale convolution kernel convolution neural network EEG classification model with the feature extraction module can achieve a maximum accuracy of 92.99% in the binary classification experiment,which is about 1 percentage point higher than before joining SSAE;In the three-category experiment,the highest can reach 90.56%,an increase of about 0.5 percentage points;For the four classifications,the accuracy rate can reach 80.51%,and the average accuracy rate is improved by about 0.3percentage points.By comparing the experimental results before and after adding the feature extraction module,it is found that the model performance has improved to a certain extent.