Emotion Recognition in Consumer Electronics Using Multimodal Deep Learning
IEEE Transactions on Consumer Electronics · 2025-08 · 71 (3) · 7636-7650
DOI: 10.1109/tce.2025.3568882 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Consumer Electronics, 2025-08. Verfasst von Wei Sun und 1 weiteren Autor:innen. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 33 Literaturangaben. Gefördert von: Jining Normal University titled “Research on Recommendation Models Based on Knowledge Graphs and Graph Neural Networks”, Inner Mongolia Academy of Social Sciences titled “Research on the Construction and Visualization of a Multilingual System for Inner Mongolia’s Outstanding Loca…, Inner Mongolia Education Science Planning titled “Research on the Innovation and Practice of the Teaching Reform of College English Courses Based on the Knowle….
Abstract (Originalsprache)
A multimodal deep learning framework is proposed to address the challenges of achieving high accuracy and real-time performance in traditional emotion recognition systems, particularly when dealing with complex emotional states and multimodal data fusion. The aim is to further the application of emotion recognition technology in consumer electronics. Initially, a combination of audio, video, and physiological signals is integrated to design a deep learning model capable of extracting and fusing features from multiple modalities. Subsequently, experiments are conducted to assess the impact of different combinations of modalities, feature selection techniques, and model architectures, with particular attention to the contribution of each modality and feature. Ablation studies further analyze the models performance across various devices and scenarios. Experimental results confirm that the proposed model significantly outperforms existing methods in emotion recognition tasks. For example, when combining audio, video, and physiological data, the model achieves an accuracy of 88.5% in emotion classification tasks, surpassing traditional single-modality approaches by over 5%. Additional device adaptability tests show that the models inference time is 250ms on smartphones and 180ms on wearable devices, demonstrating its robust real-time performance and adaptability across devices. Furthermore, comparative analysis with existing emotion recognition technologies highlights the superior robustness of the model in managing complex emotional expressions and environmental noise. These findings suggest that multimodal deep learning approaches can substantially improve emotion recognition accuracy, particularly in real-world applications, offering significant practical value.
Förderung
- Jining Normal University titled “Research on Recommendation Models Based on Knowledge Graphs and Graph Neural Networks”
- Inner Mongolia Academy of Social Sciences titled “Research on the Construction and Visualization of a Multilingual System for Inner Mongolia’s Outstanding Loca…
- Inner Mongolia Education Science Planning titled “Research on the Innovation and Practice of the Teaching Reform of College English Courses Based on the Knowle…
Zitieren
APA
Sun, W., & Li, Y. (2025). Emotion Recognition in Consumer Electronics Using Multimodal Deep Learning. IEEE Transactions on Consumer Electronics, 71(3), 7636-7650. https://doi.org/10.1109/tce.2025.3568882
Vancouver
Sun W, Li Y. Emotion Recognition in Consumer Electronics Using Multimodal Deep Learning. IEEE Transactions on Consumer Electronics. 2025;71(3):7636-7650. doi:10.1109/tce.2025.3568882
BibTeX
@article{Sun2025,
author = {Sun, Wei and Li, Yi},
title = {Emotion Recognition in Consumer Electronics Using Multimodal Deep Learning},
journal = {IEEE Transactions on Consumer Electronics},
year = {2025},
volume = {71},
number = {3},
pages = {7636--7650},
doi = {10.1109/tce.2025.3568882}
}
EUVIMED