Feature Extraction of EEG Signals Based on Local Mean Decomposition and Fuzzy Entropy

Abstract

An improved feature extraction method is proposed aiming at the recognition of motor imagined electroencephalogram (EEG) signals. Using local mean decomposition, the algorithm decomposes the original signal into a series of product function (PF) components, and meaningless PF components are removed from EEG signals in the range of mu rhythm and beta rhythm. According to the principle of feature time selection, 4[Formula: see text]s to 6[Formula: see text]s motor imagery EEG signals are selected as classification data, and the sum of fuzzy entropies of second-and third-order PF components of [Formula: see text], [Formula: see text] lead signals is calculated, respectively. Mean value of fuzzy entropy [Formula: see text] is used as input element to construct EEG feature vector, and support vector machine (SVM) is used to classify and predict EEG signals for recognition. The test results show that this feature extraction method has higher classification accuracy than the empirical mode decomposition method and the total empirical mode decomposition method.