Ponente
Descripción
Title: EEG Feature Extraction Using Wavelets for Emotion Recognition
Authors: Lic. Carlos Ángel Pérez Zapata, Dra. María Monserrat Morín Castillo, Dr. Jose Jacobo Oliveros Oliveros
Affiliation: Benemérita Universidad Autónoma de Puebla
This research leverages electroencephalography (EEG) to objectively decode emotional states, addressing limitations of subjective methods like questionnaires. By integrating EEG with machine learning (ML), the study identifies distinctive "brain signatures" linked to emotions, advancing applications in mental health diagnostics and targeted therapies.
The methodology employs the SEED-IV database, containing 62-channel EEG signals sampled at 200 Hz across four emotional categories. A rigorous preprocessing pipeline was implemented:
Bandpass filtering (1–75 Hz, 4th-order Butterworth) and notch filtering (50/60 Hz) to eliminate noise and artifacts, achieving a 15 dB SNR improvement.
Segmentation of signals into 4-second epochs.
Channel selection of 10 key electrodes using variance analysis.
Wavelet-based feature extraction, decomposing signals into approximations (A4) and details (D1–D4) to derive 20 features per epoch, supplemented by descriptive statistics.
The resulting feature sets were compiled into structured tables to train and compare multiple ML classifiers, including SVM, Random Forest, CNN, LSTM, and hybrid models.
Conclusions confirm the efficacy of wavelet decomposition in isolating discriminative neural patterns from EEG data. The preprocessing framework successfully enhanced signal quality, while variance-driven channel selection optimized computational efficiency. This pipeline provides a robust foundation for emotion recognition systems, demonstrating significant potential for clinical mental health applications. Future work will validate classifier performance across diverse emotional stimuli and explore real-time implementation.
Keywords: EEG, Wavelet Transform, Emotion Recognition, Machine Learning, SEED-IV, Feature Extraction.
