Semi-supervised Graph Convolution Neural Network based Emotion Recognition using Multimodal Biosignals
Oral presentation
Research approach
Built similarity graphs from DE and PSD features of 14 EEG channels and one channel each of EDA and PPG. Fused within-signal and cross-signal similarities before semi-supervised GCN classification of valence and arousal.
View full-size figure
Results
Subject-wise transductive evaluation on DEAP, with 32 participants and 120 labels per class. Accuracy: 97.37 ± 1.75% for valence and 97.38 ± 1.68% for arousal.
All publications