Five-channel EEG-based Emotion Recognition using CNN and LSTM Networks

This study presents an emotion recognition approach using five-channel electroencephalogram (EEG) signals analyzed through convolutional neural networks (CNN) and long short-term memory (LSTM) networks. The goal is to achieve robust binary classification (positive vs. negative emotion) using minimal electrode input to reduce computational cost. We employed five electrodes positioned in the prefrontal and temporal regions of the brain to reduce computational complexity and enhance classification accuracy. Using the SEED dataset (SJTU Emotion EEG Dataset), we applied continuous wavelet transform (CWT) to generate ۲-D images for the CNN model and wavelet packet transform (WPT) for gamma band extraction in the LSTM model. Our results show an average classification accuracy of ۹۴.۵۷±۰.۰۶% for CNN and ۷۴.۶% for LSTM. We emphasize that, despite using basic architectures, effective performance was achieved with minimal electrodes. Training time for the CNN model was ۳ minutes and for LSTM model was ۳۱ minutes that indicates the robustness of this methodology. This lightweight system has practical applications in emotion-aware technologies, mental health tools, and human-computer interaction.

کلیدواژه ها:

Emotion Recognition ، Electroencephalogram (EEG) ، Deep Learning ، Convolutional Neural Network (CNN) ، Long Short Time Memory (LSTM) ، Time-Series