BP And SVM Optimized By The Simplex Evolutionary Algorithm And Its Application In Motor Imagery EEG Signal Classification

Brain-Computer Interface(BCI)based on Motor imagery(MI)EEG is a new way of human-computer communication.The correct classification of MI EEG is the key factor to determine its performance.There are many kinds of brain-computer interface control methods,EEG signals is a characteristic signal often used in BCI system.A large amount of information about brain activity is stored in the EEG signals.By processing and studying the original EEG signals,different states of brain function can be roughly inferred,which is of great significance for the detection and treatment of cognitive disorders.There are many intelligent optimization algorithms used to analyze EEG signals,such as Genetic Algorithm,Particle Swarm Optimization,Artificial Bee Colony Algorithm,Artificial Bee Colony Algorithm,etc.But these algorithms have some limitations in the process of parameter optimization.For example,it is easy to fall into the local extremum point and depends too much on the control parameters.And the improved algorithm often adds more parameters,which increases the complexity of the algorithm.For common defects of intelligent optimization algorithms,In this paper,a new intelligent optimization algorithm — Surface-Simplex Swarm Evolution(SSSE)algorithm is introduced.The EEG signals of BCI competitions in 2003 and 2008 are taken as the research object.The mode entropy and marginal spectrum are extracted after EMD,and the AR(Auto regressive)model coefficients of the original signals are extracted,seven sets of feature vectors composed of single feature and mixed feature are used as network inputs respectively.Then,the SSSE algorithm is introduced into the parameter optimization of BP neural network and SVM respectively,and classifies BCI contest data by left-hand and right-hand.At the same time,in the classification process of BP optimized by Surface-Simplex Swarm Evolution algorithm,the time segment of data 1 is divided into seven segments,and the best recognition effect is 4-8s time period,the highest recognition rate was the combination of AR model coefficients and marginal spectrum.Comparing to experimental results of 4 subjects from data 2,the highest classification accuracy was 90.62% and the maximum Kappa value is 0.8.In the classification process of SVM optimized by Surface-Simplex Swarm Evolution algorithm,the data1 are always analyzed in 4-8s time period,and then the classification results of two sets of data of three kernel functions SVM are compared respectively.All in all,Gauss Radial basis function SVM has the best classification performance.This method realizes the accurate classification of left and right hand of MI EEG signals,and reduces the influence of control parameters on learning performance,which verifies the validity and feasibility of the algorithm in the classification and recognition of MI EEG signals.It lays a good foundation for the follow-up on-line research of BCI system based on MI EEG.