Spatio-Temporal Representation of an Electoencephalogram for Emotion Recognition Using a Three-Dimensional...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Emotion recognition is crucial for advancing human-computer interaction (HCI).
• Electroencephalogram (EEG) is a convenient and mobile tool for estimating human emotions.
• Deep neural networks (DNNs) show promise for EEG-based emotion recognition but often rely on manual feature extraction.

Purpose of the Study:

• To propose a novel method for emotion recognition using 3D Convolutional Neural Networks (3D CNNs) and efficient spatio-temporal EEG signal representation.
• To eliminate the need for handcrafted features by enabling DNNs to extract features directly from reconstructed EEG data.

Main Methods:

• Spatially reconstructed raw EEG signals from 1D time series to 2D frames based on electrode positions.
• Created a 3D EEG stream by concatenating 2D frames along the time axis.
• Applied 3D CNNs to the 3D EEG reconstructions for feature extraction and emotion classification.

Main Results:

• Achieved high classification accuracy on the DEAP dataset: 99.11% for binary valence/arousal classification and 99.73% for four-class classification.
• Demonstrated the effectiveness of the spatio-temporal representation and 3D CNNs compared to 2D CNNs and methods using handcrafted features.
• Identified optimal kernel shapes and input data configurations for the proposed method.

Conclusions:

• The proposed 3D CNN method with efficient EEG representation significantly enhances emotion recognition accuracy.
• This approach successfully leverages the spatio-temporal characteristics of EEG data, outperforming traditional methods.
• The findings suggest a more integrated and automated approach to EEG-based emotion recognition in HCI.