Fuzzy Hopfield neural network clustering for single-trial motor imagery EEG classification

Abstract

An electroencephalogram (EEG) analysis system for single-trial classification of motor imagery (MI) data is proposed in this study. Unsupervised fuzzy Hopfield neural network (FHNN) clustering, together with active segment selection and multiresolution fractal features, is used in the classification of left and right MI data. Active segment selection is used to obtain the active segment in the time-scale domain with the continuous wavelet transform (CWT) and Student’s two-sample t-statistics. The multiresolution fractal features are then extracted from the discrete wavelet transform (DWT) data by using the modified fractal dimension. Finally, FHNN clustering is used as the discriminant of multiresolution fractal features. FHNN clustering is capable of making flexible partitions of a finite data set, and it is an unsupervised and robust approach suitable for the classification of non-stationary biomedical signals. Compared with several popular supervised classifiers, FHNN clustering achieves promising results in classification accuracy.