Related Experiment Video
Updated: Jan 9, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.2K
ECG-EmotionNet: Nested Mixture of Expert (NMoE) Adaptation of ECG-Foundation Model for Driver Emotion Recognition
Summary
This study introduces ECG-EmotionNet, a novel system for recognizing driver emotions using single-channel electrocardiogram (ECG) signals in dynamic driving. The new Nested Mixture of Experts (NMoE) adaptation method achieves 70% accuracy in identifying anger, fear, sadness, and surprise.
Area of Science:
- Artificial Intelligence
- Biomedical Engineering
- Human-Computer Interaction
Background:
- Driver emotion recognition is vital for autonomous driving safety and human-autonomy interaction.
- Electrocardiogram (ECG) signals offer a promising modality for real-time emotion monitoring, especially in dynamic driving.
- Current methods often require multichannel ECG and static conditions, limiting real-world applicability.
Purpose of the Study:
- To develop a novel architecture, ECG-EmotionNet, for accurate emotion recognition from single-channel ECG signals in dynamic driving environments.
- To adapt a pre-existing ECG Foundation Model (FM) using an efficient and effective method suitable for real-time applications.
- To improve the computational efficiency and representation of ECG features for driver emotion detection.
Main Methods:
- Proposed ECG-EmotionNet architecture, adapting an ECG Foundation Model (FM) using single-channel ECG data.
- Introduced a Nested Mixture of Experts (NMoE) adaptation strategy, treating transformer layers and Short-Time Fourier Transform (STFT) features as experts.
- Fused expert embeddings via a trainable gating mechanism to capture global and local ECG features efficiently.
Main Results:
- ECG-EmotionNet achieved an average subject-level classification accuracy of 70.00% and an F1 score of 69.26%.
- The system demonstrated superior performance compared to conventional adaptation methods on a challenging driver emotion dataset.
- Successfully recognized four distinct emotional states: anger, fear, sadness, and surprise.
Conclusions:
- ECG-EmotionNet provides an effective and computationally efficient solution for driver emotion recognition in dynamic scenarios.
- The NMoE adaptation method enhances feature representation from single-channel ECG signals for improved emotion classification.
- This work advances the development of trustworthy driver monitoring systems for autonomous driving.
