Multi-category EEG signal classification developing time-frequency texture features based Fisher Vector encoding method

Abstract

Classification of electroencephalogram (EEG) signals plays an important role in the diagnosis and treatment of brain diseases in the biomedical field. Here, we introduce a different multi-category EEG signal processing technique, namely time-frequency (T-F) image representation of Gray Level Co-occurrence Matrix (GLCM) descriptors and Fisher Vector (FV) encoding for automatic classification of EEG signals. Firstly the EEG signals are converted into T-F representation by using spectrograms of Short Time Fourier Transform (STFT), which are used to obtain the T-F images. The obtained T-F images are then converted into 8-bits gray-scale images and then are divided into five sub-images corresponding to the frequency-bands of the rhythms. Then, the GLCM texture descriptors are employed to extract distinctive features which are fed into the FV encoding. Finally obtained features are fed to extreme learning machine (ELM) classifier as input for identifying abnormalities from EEG signals. The proposed method was applied to epileptic and sleep stages EEG datasets. The experimental outcomes are promising on both databases. It can be anticipated that upon its implementation in real-time practice, the proposed scheme will assist the researchers and physicians to advance the existing methods for detecting neurological diseases from EEG signals. © 2016 Elsevier B.V.