Towards Practical Emotion Recognition: An Unsupervised Source-Free Approach for EEG Domain Adaptation

TL;DR

This work tackles cross-dataset EEG emotion recognition under privacy constraints by introducing a novel source-free unsupervised domain adaptation (SF-UDA) approach. A four-stage framework combines a pre-trained feature extractor with dual classifiers and two target-adaptation modules: Dual-Loss Adaptive Regularization (DLAR) and Localized Consistency Learning (LCL), plus a PC-TTA inference strategy. Results on DEAP and SEED show state-of-the-art cross-dataset accuracies (e.g., 65.84% and 58.99%), with DLAR improving robustness to noisy pseudo-labels and LCL promoting local-consistency among reliable neighbors. The method advances privacy-preserving, cross-domain EEG emotion recognition with practical implications for affective BCIs and mental-health monitoring. It is the first SF-UDA application in EEG emotion recognition and demonstrates strong potential for real-world deployment in constrained or sensitive settings.

Abstract

Emotion recognition is crucial for advancing mental health, healthcare, and technologies like brain-computer interfaces (BCIs). However, EEG-based emotion recognition models face challenges in cross-domain applications due to the high cost of labeled data and variations in EEG signals from individual differences and recording conditions. Unsupervised domain adaptation methods typically require access to source domain data, which may not always be feasible in real-world scenarios due to privacy and computational constraints. Source-free unsupervised domain adaptation (SF-UDA) has recently emerged as a solution, enabling target domain adaptation without source data, but its application in emotion recognition remains unexplored. We propose a novel SF-UDA approach for EEG-based emotion classification across domains, introducing a multi-stage framework that enhances model adaptability without requiring source data. Our approach incorporates Dual-Loss Adaptive Regularization (DLAR) to minimize prediction discrepancies on confident samples and align predictions with expected pseudo-labels. Additionally, we introduce Localized Consistency Learning (LCL), which enforces local consistency by promoting similar predictions from reliable neighbors. These techniques together address domain shift and reduce the impact of noisy pseudo-labels, a key challenge in traditional SF-UDA models. Experiments on two widely used datasets, DEAP and SEED, demonstrate the effectiveness of our method. Our approach significantly outperforms state-of-the-art methods, achieving 65.84% accuracy when trained on DEAP and tested on SEED, and 58.99% accuracy in the reverse scenario. It excels at detecting both positive and negative emotions, making it well-suited for practical emotion recognition applications.

Figures (7)

• Figure 1: Illustration of domain adaptation strategies: (a) Traditional unsupervised domain adaptation; (b) Source-free unsupervised domain adaptation.
• Figure 2: Overview of the proposed method. The stages enclosed in red-bordered boxes are adopted from our recent UDA model imtiaz2025enhanced, while the stages enclosed in black-bordered boxes represent the original contributions of this research.
• Figure 3: Adaptation using Localized Consistency Learning.
• Figure 4: Boxplots depicting the distribution of emotion recognition accuracies on the test dataset across subjects for our proposed method and other SF-UDA methods: (a) SEED $\rightarrow$ DEAP, (b) DEAP $\rightarrow$ SEED.
• Figure 5: Confusion matrices from the ablation study: (a) SEED $\rightarrow$ DEAP, (b) DEAP $\rightarrow$ SEED.