Causal Graph Convolutional Neural Network for Emotion Recognition

Kong, Wanzeng; Qiu, Min; Li, Menghang; Jin, Xuanyu; Zhu, Li

doi:10.1109/tcds.2022.3175538

Cited by 10 publications

(12 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the adjacency matrix of the GCN graph features of EEG signals, compared with the spatial position relationship between the EEG channels [ 13 , 14 , 15 ] and functional-connectivity-based adjacency matrices [ 16 , 17 , 19 ], GC-based adjacency matrices provide potential possibility (i.e., offering more direction information) to improve the recognition accuracy for the emotion-based system. In our work, the GC-based GCN graph feature shows the superiority to the other matrices, which is consistent with the reference [ 26 ].…”

Section: Discussionsupporting

confidence: 91%

“…In this section, we compare the proposed scheme with other state-of-the-art graph features in the literature, and the comparison results are shown in Table 4 , including the support vector machine (SVM), artificial neural network (ANN), DGCNN [ 14 ], PGCNN [ 19 ], and Causal-GCN [ 26 ]. All the schemes adopt the same EEG signal division and 5-fold cross validation.…”

Section: Resultsmentioning

confidence: 99%

“…Different from functional connectivity, effective connectivity can further reveal the transmission of causal information in the brain system due to the addition of direction information. Granger causality (GC) is an effective connection measure [ 24 , 25 , 26 ]. In [ 26 ], the GC matrix was used as the adjacency matrix to construct the GCN feature, which can effectively improve the performance of the EEG recognition task.…”

Section: Introductionmentioning

confidence: 99%

“…Granger causality (GC) is an effective connection measure [ 24 , 25 , 26 ]. In [ 26 ], the GC matrix was used as the adjacency matrix to construct the GCN feature, which can effectively improve the performance of the EEG recognition task. However, the current GCN features usually take the temporal or frequency features of each frequency band of the EEG signal as the nodes, but the functional or effective matrices of the full frequency band of the EEG signals ate taken as the adjacency matrices.…”

Section: Introductionmentioning

confidence: 99%

“…The existing feature fusion schemes are commonly used in series or in weighted superpositions of EEG features of different frequency bands to effectively improve the performance of the emotion recognition task. Methods such as phase-locked-value-based graph convolutional neural networks (PGCNNs) [ 19 ], dynamical graph convolutional neural networks (DGCNNs) [ 14 ], and causal graph convolutional neural networks (Causal-GCNs) [ 26 ] use a direct cascade of the different frequency band features as the nodes, which achieves a significant improvement over the single-frequency-band features. In [ 30 ], a frequency attention mechanism was applied to adaptively allocate weights to different frequency band features, with a 0.67% average improvement over the feature summation fusion scheme.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Granger-Causality-Based Multi-Frequency Band EEG Graph Feature Extraction and Fusion for Emotion Recognition

Zhang

Chen

et al. 2022

Brain Sciences

View full text Add to dashboard Cite

Graph convolutional neural networks (GCN) have attracted much attention in the task of electroencephalogram (EEG) emotion recognition. However, most features of current GCNs do not take full advantage of the causal connection between the EEG signals in different frequency bands during the process of constructing the adjacency matrix. Based on the causal connectivity between the EEG channels obtained by Granger causality (GC) analysis, this paper proposes a multi-frequency band EEG graph feature extraction and fusion method for EEG emotion recognition. First, the original GC matrices between the EEG signals at each frequency band are calculated via GC analysis, and then they are adaptively converted to asymmetric binary GC matrices through an optimal threshold. Then, a kind of novel GC-based GCN feature (GC-GCN) is constructed by using differential entropy features and the binary GC matrices as the node values and adjacency matrices, respectively. Finally, on the basis of the GC-GCN features, a new multi-frequency band feature fusion method (GC-F-GCN) is proposed, which integrates the graph information of the EEG signals at different frequency bands for the same node. The experimental results demonstrate that the proposed GC-F-GCN method achieves better recognition performance than the state-of-the-art GCN methods, for which average accuracies of 97.91%, 98.46%, and 98.15% were achieved for the arousal, valence, and arousal–valence classifications, respectively.

show abstract