EED-CL: Extended EEG Deformer with Contrastive Learning for Robust Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion recognition from electroencephalography (EEG) signals is challenging due to complex brain activity patterns and individual differences.
• Existing methods struggle with the spatiotemporal dynamics and inter-subject variability inherent in EEG data.

Purpose of the Study:

• To develop a robust and scalable framework for accurate emotion recognition using EEG signals.
• To address the limitations of current models in handling complex EEG data and limited labeled samples.

Main Methods:

• Proposed the EED-CL framework, integrating an extended EEG-Deformer (EED) with contrastive learning (CL).
• Employed a depthwise separable convolution encoder for efficient spatiotemporal feature extraction.
• Utilized a hierarchical coarse-to-medium-to-fine (HCMFT) transformer for multiscale temporal pattern capture.
• Incorporated an attentive dense information purification (ADIP) module to reduce noise and refine features.
• Leveraged CL-based pretraining for effective feature learning with limited labeled data.

Main Results:

• The EED model outperformed conventional methods in EEG-based emotion recognition.
• The EED-CL framework demonstrated significant improvements, particularly under label-constrained conditions.
• EED-CL exhibited strong robustness against inter-subject variability and noise, ensuring stable classification with scarce labeled samples.

Conclusions:

• The EED-CL framework effectively captures multiscale spatiotemporal EEG patterns for enhanced emotion recognition.
• This approach offers a scalable and reliable solution for EEG-based emotion recognition, even with limited data and high variability.