An Emotion Recognition Embedded System using a Lightweight Deep Learning Model.

Area of Science:

• Neuroscience
• Computer Science
• Human-Computer Interaction

Background:

• Emotional state diagnosis is crucial for enhancing human-computer interaction (HCI).
• Electroencephalography (EEG) signals offer accurate and informative data for emotion recognition.
• Previous research shows correlations between EEG signals and emotional states.

Purpose of the Study:

• To develop and evaluate Convolutional Neural Network (CNN) models for emotion diagnosis using EEG signals.
• To create a lightweight and implementable model for real-time emotion recognition on embedded systems like Raspberry Pi.
• To assess model performance in both subject-dependent and subject-independent settings.

Main Methods:

• Applied three CNN models (EEGNet, ShallowConvNet, DeepConvNet) for EEG signal processing.
• Utilized baseline removal preprocessing to enhance classification accuracy.
• Evaluated models in subject-dependent and subject-independent configurations on the DEAP dataset.
• Optimized a CNN model for lightweight implementation on a Raspberry Pi for real-time emotion recognition.

Main Results:

• Achieved high classification accuracies: 99.10% for valence and 99.20% for arousal in subject-dependent settings.
• Obtained 90.76% accuracy for valence and 90.94% for arousal in subject-independent settings.
• Demonstrated real-time emotion recognition capabilities on an embedded system.

Conclusions:

• The study successfully developed a highly accurate and implementable EEG-based emotion diagnosis model.
• The lightweight CNN model is suitable for practical real-time applications in human-computer interaction.
• Results compare favorably with related works, indicating a significant advancement in the field.