Dae Hyeon Kim RESEARCH 한국어

Quantitative EEG based Emotion Recognition Using Graph Convolutional Neural Network

Oral presentation

Venue
IEIE Summer Annual Conference
Author role
First author
Year
2022.06

Authors Dae-Hyeon Kim, Young-Seok Choi

DBpia

Research approach

Decomposed EEG into five intrinsic mode functions and extracted dynamic differential entropy (DDE). Smoothed features with a moving average and classified emotions using a spectral GCN with learned electrode connections.

Emotion recognition with DDE features and a learned electrode graph
Emotion recognition with DDE features and a learned electrode graph
View full-size figure

Results

Subject-dependent evaluation on SEED with 15 participants: DDE with moving averages reached 76.13 ± 4.91% accuracy, compared with 73.39 ± 6.24% for conventional DE with moving averages.

Accuracy by EEG feature (%)
Feature Accuracy
DE 69.32 ± 12.69
DE + moving average 73.39 ± 6.24
DDE 70.37 ± 9.79
DDE + moving average 76.13 ± 4.91

Related project

EEG graph learning and compact decoding

All publications