LSTM-enhanced multi-view dynamical emotion graph representation for EEG signal recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalogram (EEG) signals contain rich information for emotion recognition.
• Existing methods often struggle to fully capture complex spatiotemporal dynamics in EEG data.

Purpose of the Study:

• To develop an LSTM-enhanced multi-view dynamic emotion graph representation model for improved EEG-based emotion recognition.
• To integrate spatial topology and temporal information from EEG signals effectively.

Main Methods:

• A two-branch model was proposed: one for dynamic multi-view graph representation learning and another for time-series information learning with memory.
• The model dynamically learns multiple graph representations and utilizes a Long Short-Term Memory (LSTM) component to retain crucial temporal information.
• Features from both branches were fused using a mean fusion operator to enhance spatiotemporal representations.

Main Results:

• Extensive subject-independent experiments were conducted on SEED, SEED-IV, and DEAP datasets.
• The proposed method demonstrated superior performance in recognizing EEG emotional signals compared to state-of-the-art approaches.

Conclusions:

• The LSTM-enhanced multi-view dynamic emotion graph representation model effectively extracts discriminative spatiotemporal features from EEG signals.
• This approach offers a promising advancement for accurate and robust emotion recognition systems.