eRAD-Fe: Emotion Recognition-Assisted Deep Learning Framework

With recent advancements in artificial intelligence technologies and human-computer interaction, strategies to identify the inner emotional states of humans through physiological signals such as electroencephalography (EEG) have been actively investigated and applied in various fields. Thus, there i...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement 2021, Vol.70, p.1-12
Main Authors: Kim, Sun-Hee, Nguyen, Ngoc Anh Thi, Yang, Hyung-Jeong, Lee, Seong-Whan
Format: Article
Language:English
Subjects:
Artificial intelligence
Covariance matrices
Data mining
Datasets
Deep learning
Demand analysis
electroencephalogram
Electroencephalography
Emotion recognition
Emotional factors
Emotions
energy threshold
Feature extraction
long short-term memory
Machine learning
multiclass common spatial pattern (CSP)
Segmentation
Speech recognition
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With recent advancements in artificial intelligence technologies and human-computer interaction, strategies to identify the inner emotional states of humans through physiological signals such as electroencephalography (EEG) have been actively investigated and applied in various fields. Thus, there is an increasing demand for emotion analysis and recognition via EEG signals in real time. In this article, we proposed a new framework, emotion recognition-assisted deep learning framework from eeg signal (eRAD-Fe), to achieve the best recognition result from EEG signals. eRAD-Fe integrates three aspects by exploiting sliding-window segmentation method to enlarge the size of the training dataset, configuring the energy threshold-based multiclass common spatial patterns to extract the prominent features, and improving emotional state recognition performance based on a long short-term memory model. With our proposed recognition-assisted framework, the emotional classification accuracies were 82%, 72%, and 81% on three publicly available EEG datasets, such as SEED, DEAP, and DREAMER, respectively.
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2021.3115195