PDGCN: A progressive dual-branch graph convolution network for EEG emotion recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalography (EEG) offers objective neural activity measurement for emotion recognition.
• Existing graph neural networks struggle with spatial topology, adaptive connectivity, and multi-scale features in EEG signals.

Purpose of the Study:

• To propose a Progressive Dual-Branch Graph Convolutional Network (PDGCN) for enhanced EEG-based emotion recognition.
• To address limitations in capturing spatial topology, adaptive connectivity, and multi-scale temporal features.

Main Methods:

• Developed PDGCN integrating progressive multi-scale temporal feature extraction and dual-branch graph convolution.
• One branch encodes intrinsic electrode topology; the other learns adaptive channel interactions via a learnable adjacency matrix.
• Validated cross-subject performance on SEED, SEED-IV, and DEAP datasets.

Main Results:

• PDGCN demonstrated consistently competitive performance against existing methods across multiple datasets.
• The framework effectively models spatial information within temporally contextualized EEG representations.
• Achieved superior emotion recognition accuracy by integrating multi-scale and multi-view feature extraction.

Conclusions:

• PDGCN provides an effective solution for EEG-based emotion recognition.
• The proposed method offers a novel perspective for multi-scale and multi-view brain signal processing.
• Highlights the potential of advanced graph convolutional networks in affective computing.