EEG discrimination using wavelet packet transform and a reduced-dimensional recurrent neural network

This paper proposes a novel reduced-dimensional recurrent neural network (NN) for electroencephalography (EEG) discrimination. Due to time-varying characteristics of EEG signals, recurrent NN is a useful approach for EEG pattern discrimination. However, when dealing with high-dimensional data, NNs usually have problems of heavy computation burden and difficulty in training. To overcome these problems, the proposed NN incorporates a dimension-reducing stage into the network structure of a recurrent probabilistic NN. Moreover, an EEG discrimination method is developed using wavelet packet transform (WPT) and the proposed NN. EEG discrimination experiments were conducted with EEG signals measured during finger movements. The experimental results of four subjects indicate that the proposed method can achieve relatively high discrimination performance.