EEG emotion recognition based on data-driven signal auto-segmentation and feature fusion

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Brain-computer interface (BCI) research utilizes pattern recognition on network connections for emotion recognition via electroencephalogram (EEG) signals.
• Current research lacks unified standards for selecting emotional signals and often overlooks brain region activation differences linked to network connectivity.
• This gap hinders the development of robust emotion recognition systems.

Purpose of the Study:

• To propose a data-driven signal auto-segmentation and feature fusion algorithm (DASF) to address limitations in EEG-based emotion recognition.
• To integrate brain functional connectivity patterns with power-distribution features for enhanced emotion classification.
• To establish a more standardized and accurate method for emotion recognition in BCI research.

Main Methods:

• Constructed dynamic brain functional networks using the Phase Locking Value (PLV) method and tucker decomposition.
• Employed data-driven methods to distinguish brain network states and automatically extract emotional signal segments.
• Utilized tensor sparse representation for feature extraction and combined power-distribution features (differential entropy, energy) with functional connectivity features for SVM classification.

Main Results:

• Achieved single-feature emotion classification accuracies of 86.57% (valence) and 87.74% (arousal).
• The proposed feature fusion method improved accuracy to 89.14% (valence) and 89.65% (arousal).
• Demonstrated superior classification performance compared to state-of-the-art methods on ERN and DEAP datasets.

Conclusions:

• The DASF algorithm effectively integrates diverse EEG signal features for improved emotion recognition.
• The data-driven approach enhances the selection and extraction of relevant emotional states from EEG data.
• This method offers a promising advancement for BCI applications requiring accurate emotion detection.