EEG Classification Based On Manifold And Swarm Intelligence Optimization

The breakthrough of EEG signal classification technology will lead to another technological revolution of human-computer interaction technology,which will have an important impact on medical care,intelligent cars and humanoid robots,and provide more effective reference for the diagnosis and treatment of brain diseases.To improve the classification accuracy of EEG signal becomes the key to EEG signal processing.Based on the theory of EEG signal processing and machine learning,this paper aims to construct a higher accuracy classification and recognition model of EEG signal.To solve the problems encountered in the process of preprocessing,feature extraction and classification of EEG signal analysis and processing,this paper combines the improved adaptive filtering algorithm,nonlinear manifold clustering dimension reduction algorithm and swarm intelligence optimization algorithm.The detailed contents and innovations are as follows.1.For EEG signal preprocessing,a preprocessing method based on the IAF(Improved Adaptive Filter)is proposed to remove the random noise and reconstruct the EEG signal.Firstly,aiming at the artifact problem of sequential random strong noise in EEG signal,the sequential random noise removal algorithm based on variable forgetting factor is proposed.Secondly,for the situation of signal loss or abnormal useless signal interference,the EEG reconstruction method based on improved kernel adaptive algorithm is proposed.The proposed EEG signal reconstruction method is based on the idea of online prediction of EEG signal.The online prediction of EEG signal is based on the EEG signal of old time points to online predict new time points by kernel adaptive algorithm.This method can effectively remove the temporal random noise and obtain the reconstructed EEG signal with smaller root mean square error.2.For EEG feature extraction,the clustering visualization method combined with manifold and two-way feature fusion method are proposed.Firstly,aiming at the unlabeled EEG data scene,different unsupervised manifolds and dimensionality reduction methods are applied to EEG data in medical treatment.Experimental results show that Landmark ISOMAP based visualization method for dimensionality reduction of EEG signal has better clustering results.The two types of sample points have a long comet like data manifold structure in low dimensions,and the boundary is obvious.Secondly,for the labeled EEG data scene,most manifold dimensionality reduction methods do not make full use of label information in training data.The proposed method makes full use of the label information of discriminant features by combining nonlinear dimension reduction and linear dimension reduction with supervised algorithm.The proposed two-way feature fusion algorithm can more effectively extract the nonlinear inherent manifold features,linear principal component features and discrimination features including label information in the segmented EEG data.Experiments show that the average classification accuracy of the proposed feature fusion method is improved3.For EEG signal classification and recognition,the ELM method has IRSLFN(Ill-conditioned Random Single-Hidden-Layer Feedforward Network)problem with random initial weight and the limitation of activation function.To solve this problem,the new swarm intelligence algorithms and various bionic strategies are proposed to optimize the initialization weight,bias and activation function of different ELM algorithms.Firstly,the EEG classification method based on spiking swarm intelligent optimization algorithm is proposed.The neuron individuals in the algorithm can decide whether to release the spiking according to their own energy accumulation.Compared with the commonly used activation function,the algorithm can improve the ability of network to process information.Secondly,aiming at the IRSLFN problem of random initialization weight and bias,three improved bionic swarm intelligence algorithms are proposed for EEG classification.The first algorithm optimizes the initial weight and bias by swarm intelligence algorithm,making it closer to the optimal solution of global optimization problem.In order to avoid misjudgment of the best individual,the top several better individuals are selected according to the fitness value in each iteration and voted for the final decision.The appropriate regularization factors are introduced in different layers to make the information processing sparser.Finally,a series of experiments are carried out to verify the effectiveness of the proposed algorithm.