Feature selection and integration is one of the crucial approaches to improving the emotion decoding accur-acy of electroencephalogram(EEG)signals.However,current methods often neglect the implicit information of the in-trinsic data structure in EEG signals.Herein,a multitask feature integration method is proposed based on affinity propagation clustering.This method uses the L2,1-norm constraint to select sparse features and uses graph Laplacian reg-ularization to maintain potential relationships among different subclasses.In case of not disclosing real sample labels,the method has effectively integrated the spatial topology information of brain networks and differential entropy inform-ation in the subtask space,providing features with higher emotional characterization ability for the emotional decoding of high-accuracy EEG signals.The analytic results on DEAP and SEED datasets and the dataset of the laboratory show that the proposed method can markedly improve the decoding accuracy of EEG emotional decoding.
维普
万方数据