SAE+LSTM: A New Framework for Emotion Recognition From Multi-Channel EEG.

Area of Science:

• Neuroscience
• Robotics
• Artificial Intelligence

Background:

• Brain-inspired robots require effective human interaction capabilities.
• Automatic emotion recognition from electroencephalogram (EEG) is crucial for enhancing human-robot collaboration.
• Current methods face challenges in accurately interpreting complex EEG signals for emotion detection.

Purpose of the Study:

• To develop a novel framework for multi-channel EEG-based emotion recognition.
• To improve the accuracy and robustness of emotion classification for human-robot interaction.
• To leverage deep learning models for advanced EEG signal decomposition and analysis.

Main Methods:

• A novel framework combining a linear EEG mixing model and an emotion timing model was proposed.
• Stack AutoEncoder (SAE) was utilized to decompose EEG source signals.
• Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) was employed for emotion timing analysis.
• The framework was validated using the DEAP dataset.

Main Results:

• The proposed framework achieved a mean accuracy of 81.10% for valence and 74.38% for arousal.
• Contextual correlations in EEG feature sequences significantly improved classification accuracy.
• The framework demonstrated superior performance compared to conventional approaches in multi-channel EEG emotion recognition.

Conclusions:

• The novel EEG-based emotion recognition framework effectively improves classification accuracy.
• The integration of SAE and LSTM-RNN offers a powerful approach for analyzing complex EEG data.
• This advancement holds significant potential for enhancing human-robot interaction through more intuitive and accurate emotional understanding.