Classification of EEG signals based on time-frequency analysis and spiking neural network

Electroencephalogram (EEG) is one of the most effective and essential tools for analyzing and diagnosing epilepsy. However, there is a challenging task for detecting seizures from EEG signals, which is due to the non-stationary nature of EEG signals. This paper proposes a novel automatic EEG signal recognition method to assist epilepsy detection. Specifically, the multi-wavelet basis function (MWBF) expansion method is first adopted to construct a time-varying autoregressive (TVAR) model of EEG signals, and a robust ultra-orthogonal least squares (UOLS) algorithm aided by derivative information of EEG signals is then utilized for model structure detection; besides, the power spectral density (PSD) estimation method is applied to extract high-resolution time-frequency features; particularly, to fully exploited the spatiotemporal information of the extracted features, features were fed into the spiking neural networks (SNN) for classification. Experimental results on a widely-used benchmark dataset show that proposed methods outperform other related methods in terms of classification performance.