Abstract

This study investigated the effects of different transforming techniques on the Electroencephalography (EEG) signal analysis in classifying the emotional SEED-V EEG dataset. Using different traditional and advanced transformation methods, including the Fourier Transform (FT), Short-Time Fourier Transform (STFT), Spectrogram Transform, Spectrogram Contrast Limited Adaptive Histogram Equalization (SCLAHE) and Hilbert Transform, the study examines the ability of these preprocessing approaches to clarify and extract more features of the EEG data by transforming them from Real numbers to the Complex numbers. Accordingly, the results enhance the processing of the spectrogram pictures, and further, SCLAHE and the Hilbert Transform significantly enhance the classification of the data. In particular, the Hilbert Transform was able to extract instantaneous phase and amplitude information, allowing for a better understanding of brain connection and synchronization, seeing a 49% increase in the Feedforward Neural Networks (FFNN) classification accuracy compared to using the main EEG signal result as a benchmark. Also, for the CNN, the accuracy improves by over 100% compared to the STFT benchmark when applying SCLAHE preprocessing on data. The results indicate a significant advantage behind the use of multiple preprocessing methods, allowing for complex interaction identification within the EEG results, offering a way for further research to develop this idea and combine it with computational models for a deeper insight into brain operations.

Keywords:Electroencephalography (EEG); Fourier Transform (FT); Short-Time Fourier Transform (STFT); Contrast Limited Adaptive Histogram Equalization (CLAHE); Hilbert Transform; Spectrogram Contrast Limited Adaptive Histogram Equalization (SCLAHE)

Abbreviations:EEG: Electroencephalography; FT: Fourier Transform; STFT: Short-Time Fourier Transform, CLAHE: Contrast Limited Adaptive Histogram Equalization, CNN: Convolutional Neural Networks, FFNN: Feedforward Neural Networks