SDH-LGCNet: Static–Dynamic Hybrid Spatio-Temporal Graph Learning With Factorized Spatial Priors for EEG Emotion Recognition

Electroencephalogram (EEG) signals can directly reflect neural activity and are an important source of information for affective computing. However, using only static or dynamic graphs makes it difficult to accommodate both stable sensor-space priors and input-adaptive dependencies. Electrode-level local modeling also lacks global integration and feedback guided by region and lateralization priors. We therefore propose SDH-LGCNet, a static-dynamic hybrid three-dimensional (3D) graph convolutional network with factorized prior-guided local-global collaborative learning. Within local temporal windows, the static-dynamic hybrid 3D graph convolution module (SDH-G3D) constructs a static spatio-temporal prior graph and an input-adaptive sparse dynamic graph. Through residual gating and multi-order propagation, SDH-G3D learns local spatio-temporal representations. The factorized prior-guided local-global collaborative learning module (LGCL) encodes five scalp-topographic regions and left, midline, and right lateralization positions as factorized prior tokens. Local-to-global aggregation, token context learning, and global-to-local feedback jointly integrate local details with global context. On the public Shanghai Jiao Tong University Emotion EEG Dataset (SEED) and Database for Emotion Analysis using Physiological Signals (DEAP), SDH-LGCNet achieves competitive recognition performance under both subject-dependent and subject-independent settings. Ablation experiments on SEED further validate the effectiveness of static-dynamic graph fusion, cross-time propagation, and joint modeling of region and lateralization priors. The approach provides a modeling framework for the collaborative learning of local spatio-temporal relations and global spatial information in EEG.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11366891/v1
Primary Topic
Emotion and Mood Recognition
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

SDH-LGCNet: Static–Dynamic Hybrid Spatio-Temporal Graph Learning With Factorized Spatial Priors for EEG Emotion Recognition

Yindong Dong, Yangfan Li, Yubo Xing, Xiule Wang et al.
3 links
Research Square
Emotion and Mood Recognition
preprint

SDH-LGCNet: Static–Dynamic Hybrid Spatio-Temporal Graph Learning With Factorized Spatial Priors for EEG Emotion Recognition

Yindong Dong, Yangfan Li, Yubo Xing, Xiule Wang, Tian Zhao, Jie He, Yiming Tang
preprint en

Abstract

Electroencephalogram (EEG) signals can directly reflect neural activity and are an important source of information for affective computing. However, using only static or dynamic graphs makes it difficult to accommodate both stable sensor-space priors and input-adaptive dependencies. Electrode-level local modeling also lacks global integration and feedback guided by region and lateralization priors. We therefore propose SDH-LGCNet, a static-dynamic hybrid three-dimensional (3D) graph convolutional network with factorized prior-guided local-global collaborative learning. Within local temporal windows, the static-dynamic hybrid 3D graph convolution module (SDH-G3D) constructs a static spatio-temporal prior graph and an input-adaptive sparse dynamic graph. Through residual gating and multi-order propagation, SDH-G3D learns local spatio-temporal representations. The factorized prior-guided local-global collaborative learning module (LGCL) encodes five scalp-topographic regions and left, midline, and right lateralization positions as factorized prior tokens. Local-to-global aggregation, token context learning, and global-to-local feedback jointly integrate local details with global context. On the public Shanghai Jiao Tong University Emotion EEG Dataset (SEED) and Database for Emotion Analysis using Physiological Signals (DEAP), SDH-LGCNet achieves competitive recognition performance under both subject-dependent and subject-independent settings. Ablation experiments on SEED further validate the effectiveness of static-dynamic graph fusion, cross-time propagation, and joint modeling of region and lateralization priors. The approach provides a modeling framework for the collaborative learning of local spatio-temporal relations and global spatial information in EEG.

Research Square
Anhui Agricultural University (CN), Hefei University of Technology (CN)
National Natural Science Foundation of China
Emotion and Mood Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon