Application Of EEG Signal Pattern Recognition And Limb Rehabilitation Training Based On Motor Imagination

With the intersecting development of brain science and engineering science,the technology of constructing interaction with external devices by decoding human EEG signals is developing rapidly.The main purpose of this thesis is to study the pattern recognition and application in physical rehabilitation technology of EEG signals based motor imagery.This thesis makes an introduction and study on the preprocessing,feature engineering,pattern recognition and application of EEG signals.Based on the newly proposed squirrel search algorithm,the improved squirrel search algorithm(GOSSA)and the multi-objective squirrel search algorithm(MOSSA)are proposed.The GOSSA is applied to optimize the parameters of variational mode decomposition(VMD),and then it can be applied to the removal of ocular artifacts from EEG signals.The multiobjective squirrel search algorithm is used to reduce feature redundancy by selecting features of EEG signals.Support Vector Machine(SVM)is selected as the classification and recognition model in this thesis.In view of the parameter problem of support vector machine,the offline data and GOSSA were combined to optimize the parameters to get the optimal parameters,and then the classifier model was trained.Finally,the EEG signals based on motor imagery were applied to the limbs rehabilitation robot,and the training classifier was used to identify and control the robot online,and the basic functional test was completed.The main contents of the thesis are as follows:1)The thesis introduces the present methods and basic steps of EEG signal preprocessing.Opposition-based learning is introduced into squirrel search algorithm,and the traction of global optimal solutions is strengthened.The improved squirrel search algorithm(GOSSA)is applied to the removal of ocular artifacts from EEG signals by combining with variational mode decomposition(VMD).The method is verified by the test performance of a data set with ocular artifacts from EEG signals2)It introduces time and frequency domain of EEG signals,the calculation method of compute nonlinear feature extraction of EEG signals is also introduced.In order to improve the speed and accuracy of pattern recognition and remove redundant features,this thesis proposes a multi-objective squirrel search algorithm for feature selection by combining squirrel search algorithm with grid method.3)The data acquisition of the EEG signals based on motor imagery from subjects is realized by BP EEG acquisition equipment,and then extracting features.Multiobjective optimization algorithm is used for feature selection to obtain feature set.The feature set is inputting to the model which is combining GOSSA and SVM,and then the optimal parameters can be obtained to train the classification and recognition model.Carrying out real time data process when obtaining real-time acquisition of the EEG signals based on motor imagery from subjects,and then classifying the data in time through the trained classification model.Real-time classification of EEG signals based on motor imagery is applied to the limbs rehabilitation robot to control the motion pattern.4)In Appendix I and II,the improved squirrel search algorithm and multi-objective squirrel search algorithm proposed in this thesis are tested.It can be shown from the performance that the two algorithms proposed in this thesis have better property of convergence by comparing the results with similar algorithms in typical calculatingexamples.Variational Mode Decomposition(VMD)and improved squirrel search algorithm are combined to remove the ocular artifact.The results of an example show that this method is effective.The squirrel search algorithm is extended from single objective optimization to multi-objective optimization,so a multi-objective squirrel search algorithm is proposed.The multi-objective squirrel search algorithm is used to select EEG signal features,reduce feature redundancy,and improve the accuracy of classification results.First,offline data was used to train the classification model,and then the model was used for real-time recognition of EEG signals.The recognition result is used to control the lower limb rehabilitation robot independently developed by the research group,and its online classification accuracy rate reaches about 70%.