[Using electroencephalogram for emotion recognition based on filter-bank long short-term memory networks].

Area of Science:

• Neuroscience
• Computer Science
• Affective Computing

Background:

• Emotion recognition is crucial for human-computer interaction.
• Electroencephalogram (EEG) signals offer insights into internal emotional states.
• Affective brain-computer interfaces (BCIs) leverage EEG for enhanced interaction.

Purpose of the Study:

• To evaluate state-of-the-art feature extraction and classification methods for EEG-based emotion recognition.
• To introduce and validate a novel Filter-bank long short-term memory network (FBLSTM) model.
• To optimize emotion recognition accuracy using a 4-second time window and block-wise K-fold cross-validation.

Main Methods:

• Systematic evaluation of feature extraction and classification techniques on the DEAP dataset.
• Implementation of block-wise K-fold cross-validation to prevent data leakage.
• Development and application of a Filter-bank long short-term memory network (FBLSTM) model utilizing differential entropy features.

Main Results:

• A 4-second time window was identified as optimal for EEG signal sampling.
• The FBLSTM model achieved high average accuracies: 78.8% for valence, 78.4% for arousal, and 70.3% for the valence-arousal plane.
• Demonstrated superior classification accuracy compared to existing emotion recognition studies.

Conclusions:

• The proposed FBLSTM model offers a novel and effective method for EEG-based emotion recognition.
• This research advances the field of affective brain-computer interactions.
• Accurate emotion recognition using EEG signals holds significant potential for future human-computer interaction applications.