Multimodal-informed emotional stress recognition based on ECG
- Publication Type:
- Thesis
- Issue Date:
- 2026
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Emotional stress recognition using physiological signals has attracted increasing attention due to its potential applications in mental health monitoring and wearable systems. Among various biosignals, electrocardiogram (ECG) offers strong practicality for continuous monitoring, but its discriminative capability is limited when used as a single modality under realistic conditions. While multimodal approaches can improve performance by leveraging complementary information, their reliance on multiple sensors constrains practical deployment.
This thesis proposes a training–inference paradigm that enhances ECG-based stress recognition while preserving deployability. Complementary physiological signals, including electrodermal activity (EDA) and electromyography (EMG), are utilized during training to provide enriched supervision, whereas ECG is retained as the sole modality during inference. A unified signal processing pipeline and a teacher–student knowledge distillation framework are developed to transfer multimodal knowledge into a compact ECG-based model.
The proposed approach is evaluated on multiple public datasets and a self-collected multimodal dataset under both binary and multi-class settings. Results demonstrate consistent improvements in classification performance, convergence stability, and robustness compared with ECG-only baselines.
Overall, this study shows that multimodal-informed training with knowledge distillation provides an effective solution to improve ECG-based stress recognition under practical deployment constraints.
Please use this identifier to cite or link to this item: