Scholarlink

<p>Affective computing plays a vital role in understanding and influencing human behavior, decision-making, and task performance. In this context, affect analysis in a three-dimensional (3D) space—defined by Valence, Arousal, and Dominance (VAD) is crucial but challenging due to the dynamic and complex nature of emotions. This study proposes a hybrid deep learning architecture for emotion analysis in the 3D VAD space, integrating a Deep Convolutional Generative Adversarial Network (DCGAN) with inception-V3 and MobileNet, the two most powerful convolutional neural networks. Electroencephalogram (EEG) signals were selected for their non-invasive nature, cost-effectiveness, and high temporal resolution. We have utilized the scalograms—time–frequency representations of EEG signals as inputs. Experiments were conducted on the benchmark DEAP database, ensuring reliable affective computing research. The proposed hybrid framework achieved high classification accuracy, with 75.49% using DCGAN-Inception-V3 and 75.30% with DCGAN-MobileNet, outperforming many existing studies, particularly in 3D VAD analysis. In discrete emotion classification, the DCGAN-MobileNet model consistently outperformed Inception-V3 with test accuracies of 96.41% and 95.76% for four-class and six-class classification, respectively. These findings highlight the effectiveness of integrating scalograms with deep CNN models and demonstrate the framework’s potential to enhance classification accuracy beyond traditional two-dimensional approaches, capturing more nuanced emotional states.</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

Hybrid DCGAN-CNN Architecture for EEG Based Emotion Recognition in 3D Valence-Arousal-Dominance Space

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Gyanendra K. Verma

摘要

Affective computing plays a vital role in understanding and influencing human behavior, decision-making, and task performance. In this context, affect analysis in a three-dimensional (3D) space—defined by Valence, Arousal, and Dominance (VAD) is crucial but challenging due to the dynamic and complex nature of emotions. This study proposes a hybrid deep learning architecture for emotion analysis in the 3D VAD space, integrating a Deep Convolutional Generative Adversarial Network (DCGAN) with inception-V3 and MobileNet, the two most powerful convolutional neural networks. Electroencephalogram (EEG) signals were selected for their non-invasive nature, cost-effectiveness, and high temporal resolution. We have utilized the scalograms—time–frequency representations of EEG signals as inputs. Experiments were conducted on the benchmark DEAP database, ensuring reliable affective computing research. The proposed hybrid framework achieved high classification accuracy, with 75.49% using DCGAN-Inception-V3 and 75.30% with DCGAN-MobileNet, outperforming many existing studies, particularly in 3D VAD analysis. In discrete emotion classification, the DCGAN-MobileNet model consistently outperformed Inception-V3 with test accuracies of 96.41% and 95.76% for four-class and six-class classification, respectively. These findings highlight the effectiveness of integrating scalograms with deep CNN models and demonstrate the framework’s potential to enhance classification accuracy beyond traditional two-dimensional approaches, capturing more nuanced emotional states.