Abstract:Emotion recognition has significant potential in healthcare and affect-sensitive systems such as brain-computer interfaces (BCIs). However, challenges such as the high cost of labeled data and variability in electroencephalogram (EEG) signals across individuals limit the applicability of EEG-based emotion recognition models across domains. These challenges are exacerbated in cross-dataset scenarios due to differences in subject demographics, recording devices, and presented stimuli. To address these issues, we propose a novel approach to improve cross-domain EEG-based emotion classification. Our method, Gradual Proximity-guided Target Data Selection (GPTDS), incrementally selects reliable target domain samples for training. By evaluating their proximity to source clusters and the models confidence in predicting them, GPTDS minimizes negative transfer caused by noisy and diverse samples. Additionally, we introduce Prediction Confidence-aware Test-Time Augmentation (PC-TTA), a cost-effective augmentation technique. Unlike traditional TTA methods, which are computationally intensive, PC-TTA activates only when model confidence is low, improving inference performance while drastically reducing computational costs. Experiments on the DEAP and SEED datasets validate the effectiveness of our approach. When trained on DEAP and tested on SEED, our model achieves 67.44% accuracy, a 7.09% improvement over the baseline. Conversely, training on SEED and testing on DEAP yields 59.68% accuracy, a 6.07% improvement. Furthermore, PC-TTA reduces computational time by a factor of 15 compared to traditional TTA methods. Our method excels in detecting both positive and negative emotions, demonstrating its practical utility in healthcare applications. Code available at: <a class="link-external link-https" href="https://github.com/RyersonMultimediaLab/EmotionRecognitionUDA" rel="external noopener nofollow">this https URL</a>

CFDA-CSF: A Multi-modal Domain Adaptation Method for Cross-subject Emotion Recognition

Multi-view Domain-Adaptive Representation Learning for EEG-based Emotion Recognition

A Efficient Multimodal Framework for Large Scale Emotion Recognition by Fusing Music and Electrodermal Activity Signals

Multi-source joint domain adaptation for cross-subject and cross-session emotion recognition from electroencephalography

Multi-Source EEG Emotion Recognition via Dynamic Contrastive Domain Adaptation

FMLAN: A novel framework for cross-subject and cross-session EEG emotion recognition

MS-MDA: Multisource Marginal Distribution Adaptation for Cross-Subject and Cross-Session EEG Emotion Recognition

Spectral-Spatial Attention Alignment for Multi-Source Domain Adaptation in EEG-Based Emotion Recognition

Easy Domain Adaptation for cross-subject multi-view emotion recognition

Multi-Source Co-adaptation for EEG-Based Emotion Recognition by Mining Correlation Information

Hybrid transfer learning strategy for cross-subject EEG emotion recognition

Gusa: Graph-based Unsupervised Subdomain Adaptation for Cross-Subject EEG Emotion Recognition

Multi-Source and Multi-Representation Adaptation for Cross-Domain Electroencephalography Emotion Recognition

Subject-Based Domain Adaptation for Facial Expression Recognition

Enhanced Cross-Dataset Electroencephalogram-based Emotion Recognition using Unsupervised Domain Adaptation

Electroencephalogram Emotion Recognition Based on Empirical Mode Decomposition and Optimal Feature Selection.

Multi-Source Domain Adaptation with Transformer-based Feature Generation for Subject-Independent EEG-based Emotion Recognition

Multi-source Selective Graph Domain Adaptation Network for cross-subject EEG emotion recognition

MGFKD: A semi-supervised multi-source domain adaptation algorithm for cross-subject EEG emotion recognition

EEG-based cross-subject emotion recognition using multi-source domain transfer learning

MS-FRAN: A Novel Multi-Source Domain Adaptation Method for EEG-Based Emotion Recognition