TPRO-NET: an EEG-based emotion recognition method reflecting subtle changes in emotion.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based emotion recognition is crucial for human-computer interaction and healthcare.
• Existing methods often oversimplify emotion dimensions (Valence-Arousal-Dominance) into high/low categories, missing nuanced emotional states.
• Challenges remain in EEG feature design and transformer efficiency for accurate emotion classification.

Purpose of the Study:

• To develop an advanced neural network, TPRO-NET, for more precise emotion recognition.
• To improve the classification of subtle emotional variations within the Valence-Arousal-Dominance model.
• To address limitations in current EEG feature extraction and transformer architectures.

Main Methods:

• TPRO-NET utilizes differential entropy and enhanced differential entropy features as input.
• The network incorporates convolutional layers and improved transformer encoders for emotion classification.
• Experiments were conducted on the DEAP (8 classes) and DREAMER (5 classes) datasets.

Main Results:

• TPRO-NET achieved high accuracy rates on both datasets for subject-dependent experiments.
• DEAP dataset results: 97.63% (Valence), 97.47% (Arousal), 97.88% (Dominance).
• DREAMER dataset results: 98.18% (Valence), 98.37% (Arousal), 98.40% (Dominance).
• TPRO-NET outperformed other advanced emotion recognition methods.

Conclusions:

• TPRO-NET demonstrates superior performance in EEG-based emotion recognition compared to existing approaches.
• The proposed network effectively captures subtle emotional variations, advancing the Valence-Arousal-Dominance model.
• TPRO-NET offers a promising solution for enhanced emotion recognition in various applications.