A Comparative Study of Time–Frequency Features Based Spatio-temporal Analysis with Varying Multiscale Kernels for Emotion Recognition from EEG

This paper studies EEG-based emotion recognition using a spatio-temporal analysis pipeline built on time–frequency features. Using the DEAP dataset, the authors apply downsampling, bandpass filtering, segmentation, trimming, labeling, and common reference averaging as preprocessing.

Features are extracted via Continuous Wavelet Transform (CWT) with Morlet wavelets, followed by the computation of differential entropy, wavelet energy, cross-correlation, and phase locking value (PLV). A 3D CNN with multiscale kernels (3×3×3, 5×5×5, 7×7×7) is used for classification.

The best configurations achieve training accuracies up to 98.86% (arousal) and 98.97% (valence), and test accuracies up to 96.13% and 96.19% for specific feature–kernel combinations. Validation on the SEED dataset further confirms generalization, highlighting effective choices of features and kernel sizes for EEG emotion models.

Recommended citation: Md Raihan Khan, Airin Akter Tania, and Mohiuddin Ahmad, "A comparative study of time–frequency features based spatio-temporal analysis with varying multiscale kernels for emotion recognition from EEG," Biomedical Signal Processing and Control, Elsevier, 2025.
Download Paper