Study On Classification Of Motor Imagery Electroencephalograph Signals Based On Artificial Neural Network

Brain computer interface has great significance in human computer interaction field.While motor imagery based BCI has great research value in aspects of medical,military and civil application.This mode is mainly focusing on feature extraction and classification of EEG signals under various motor imagery tasks which leads to achieve the purpose of recognizing the motor intention of human brain.Traditional researches mainly did digging in feature in one single physical domain or simply combined with multiple physical domains' features.They didn't take the time,frequency and spatial domain as a whole feature,thus their recognition rates were limited so as the generalization ability of the model.Aiming at these problems,the study started from the modeling of simple two-class motor imagery EEG signal classification.After the completion of simple model,the model was improved into multi-class recognition model.For modeling the two-class motor imagery EEG signal classifier,this study proposed an improved multilayer perceptron model for the feature fusion and classification of EEG features in time domain,frequency domain and spatial domain.By modifying multilayer perceptron structure,three domains' features were fused and classified by this structure.This model performed well in recognition rate.Based on the modified model,a multi-person multi-class recognition model was desired.Succeeding from the two-class model,the feature map was proposed as feature extraction method.By analyzing the results of the same multi-class motor imagery EEG signal dataset,the model of this study structured was better than other methods.Due to the small number of samples in public datasets,the study established a selfbuilt multi-class motor imagery EEG signal database to supplement and validate the model's generalization ability and classification rate with a total duration of 763 minutes.Using the model above to process the self-built database,it showed that the model had good generalization ability to different subjects and different motor imagery tasks,and it in turn validated the validity of this database.In conclusion,this study has established a multi-people multi-class motor imagery based EEG signal classification model,this model has the ability to achieve high recognition rate and generalization based on the public dataset.And study has built a 44-people dataset which has testify the validity of proposed model.