Time–Frequency EEG Emotion Recognition on DEAP & SEED

This project contains the full implementation for the paper:

“A comparative study of time–frequency features based spatio-temporal analysis with varying multiscale kernels for emotion recognition from EEG” (BSPC, Elsevier, 2025).

Using the DEAP (and validation on SEED), the repository provides:

• A complete preprocessing pipeline: downsampling, band-pass filtering, segmentation, common reference averaging, and labeling.
• Feature extraction notebooks for:
  • Differential Entropy (DE)
  • Wavelet Energy (WE)
  • Cross-Correlation (XCOR)
  • Phase Locking Value (PLV)
• Construction of spatio-temporal EEG tensors for 3D-CNN models with different multiscale kernels (3×3×3, 5×5×5, 7×7×7).
• Training code (model-training.ipynb and variants) that reproduces the main experimental results, including accuracies above 96% for both arousal and valence.
• Figure generation scripts used in the manuscript.

The repo is meant as a reproducible reference implementation for researchers working on EEG-based emotion recognition with time–frequency features and multiscale 3D-CNNs.