Scholarlink

<p>Emotion recognition is vital in healthcare, human-computer interaction, and affective computing. However, existing unimodal and multimodal approaches often suffer from low accuracy, high computational costs, and susceptability to noisy data. This study introduces a lightweight, real-time multimodal emotion recognition framework that integrates Electroencephalogram (EEG) and audio signals with edge computing. The proposed system leverages advanced preprocessing, deep feature extraction using a modified ResNet-152, a Feedback Attention Network, and optimized fusion via the Osprey Optimization Algorithm. Classification is performed using a novel Hybrid Dynamic Graph-Attention Residual Network (HDGAR-Net), further refined with a Snow Ablation Optimizer. Experiments on DEAP and RAVDESS datasets demonstrate the model’s superior performance, achieving 99.75% accuracy, 99.72% recall, and 99.71% F1-score. These results make it suitable for real-world deployment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Edge-enabled Hybrid Dynamic Graph-Attention Residual Network for enhanced multimodal emotion recognition

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • M. Bharaneedharan,
  • G. Kavya,
  • T. Annalakshmi,
  • B. R. Tapas Bapu

摘要

Emotion recognition is vital in healthcare, human-computer interaction, and affective computing. However, existing unimodal and multimodal approaches often suffer from low accuracy, high computational costs, and susceptability to noisy data. This study introduces a lightweight, real-time multimodal emotion recognition framework that integrates Electroencephalogram (EEG) and audio signals with edge computing. The proposed system leverages advanced preprocessing, deep feature extraction using a modified ResNet-152, a Feedback Attention Network, and optimized fusion via the Osprey Optimization Algorithm. Classification is performed using a novel Hybrid Dynamic Graph-Attention Residual Network (HDGAR-Net), further refined with a Snow Ablation Optimizer. Experiments on DEAP and RAVDESS datasets demonstrate the model’s superior performance, achieving 99.75% accuracy, 99.72% recall, and 99.71% F1-score. These results make it suitable for real-world deployment.