Emotion Recognition Model of EEG Signals Based on Double Attention Mechanism

Area of Science:

• Affective computing
• Human-computer interaction
• Neuroscience

Background:

• Emotions significantly impact human cognition, decision-making, and communication.
• Accurate emotion recognition from brain signals is a key challenge in affective computing and HCI.
• Existing deep learning models struggle with feature extraction and accuracy for emotion recognition.

Purpose of the Study:

• To propose a novel multi-channel automatic classification model for emotion EEG signals.
• To address limitations in feature extraction and accuracy of current deep learning models.
• To enhance emotion recognition performance using advanced deep learning architectures.

Main Methods:

• Developed the DACB model, integrating dual attention mechanisms, convolutional neural networks (CNNs), and bidirectional long short-term memory (BiLSTM) networks.
• Employed SE attention modules within CNNs to learn channel feature importance.
• Utilized dot product attention mechanisms to learn spatial and temporal feature importance.

Main Results:

• Achieved high single-shot validation accuracy: 99.96% on SEED-IV and 87.52%-90.06% on DREAMER datasets.
• Demonstrated strong 10-fold cross-validation accuracy: 99.73% on SEED-IV and 84.26%-85.40% on DREAMER datasets.
• Outperformed existing models in emotion classification tasks on benchmark datasets.

Conclusions:

• The DACB model exhibits high accuracy and strong generalization ability for EEG-based emotion classification.
• The proposed model offers a promising new direction for research in EEG signal recognition.
• This work advances the field of emotion recognition through improved deep learning techniques.