The multiscale 3D convolutional network for emotion recognition based on electroencephalogram.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Emotion recognition from electroencephalogram (EEG) is a key area in brain-computer interfaces (BCI).
• Convolutional neural networks (CNNs) offer advantages in automatic feature extraction for EEG analysis compared to traditional machine learning.
• Multiscale kernels in CNNs can enhance non-linear expression capabilities.

Purpose of the Study:

• To propose a 3D convolutional neural network (CNN) model utilizing multiscale convolutional kernels for enhanced EEG-based emotion recognition.
• To accurately classify four distinct emotional states based on valence and arousal levels derived from EEG signals.
• To evaluate the proposed model's effectiveness on established DEAP and SEED-IV datasets.

Main Methods:

• Development of a 3D CNN architecture incorporating multiscale convolutional kernels.
• Selection of optimal time window data from EEG signals for emotion classification.
• Training and validation of the model using DEAP and SEED-IV datasets.

Main Results:

• The proposed emotion recognition network (ERN) model achieved accuracies of 95.67% on the DEAP dataset.
• The ERN model attained an accuracy of 89.55% on the SEED-IV dataset.
• The experimental results validate the efficacy of the multiscale kernel approach in EEG-based emotion recognition.

Conclusions:

• The proposed 3D CNN model with multiscale kernels demonstrates high performance in recognizing emotional states from EEG.
• This approach shows significant potential for improving emotional feedback and user experience in BCI applications.
• Further research can explore the integration of this model for real-time emotion-aware BCI systems.