No abstract

Short-time Fourier transform is a time-frequency method commonly used to analyze signals, particularly in EEGs. It has shown acceptable results for the identification of different actions, such as sleep disorders, epilepsy, and others, and in applications as brain-computer interfaces. However, the selection of short time Fourier transform parameters is not a trivial task, as the variability of these directly affects the resolution spectrogram, from which features are extracted to determine the constructed models in the classification stage. In this paper, experiments for determining STFT parameters such as window type and length, and overlapping are explored. As a case study, an EEG epilepsy database is used to identify healthy people versus patients suffering epileptic seizures, finding that the parameters modify the spectrogram visualization in terms of time/frequency and classification. Based on these experiments, it was concluded that the proposed strategy supports the correct selection of parameters that positively impact the accuracy of the results obtained.

EEG analysis of epileptic seizure is a nonstationary and changing process; these EEGs contain multiple frequencies and use only conventional methods based on frequency or time which limit their analysis. In this paper, texture representation of Gabor filter response based on a spectrogram applying a STFT is proposed to classify epileptic and healthy states. As initial part of our research, energy and entropy were extracted from the Gabor filters response; these features and statistical values were employed to train Support Vector Machines and multilayer Perceptron classifiers.