Scholarlink

<p>Autism spectrum disorder (ASD) children face communication, behavior, and social challenges, making it challenging for caregivers to provide constant supervision and address unusual behaviors. To address this, the proposed system monitors the emotions of an ASD child by analyzing their facial expressions. When the system detects emotions such as fear, sadness, or anger, it generates a spoken avatar of the child's caregiver, to provide immediate support and intervention. The proposed system comprises two main components working concurrently: real-time facial emotion detection and talking avatar generation. For the real-time emotion detection component, an optimized version of the Xception deep-learning architecture is introduced, utilizing a modified Hippopotamus Optimization algorithm (MHO). The optimized Xception model demonstrated superior performance across three benchmark datasets. The model achieved impressive accuracies of 99.61% on Extended CK + 48, 82.68% on FER2013, and 97.18% on autistic children emotions dataset. Additionally, the results demonstrate that MHO algorithm significantly enhances the Xception model's performance and the overall real-time emotion detection system. </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

Optimized deep learning in a metaverse environment for autistic child support

  • 期刊论文

  • 发表日期：

    2025年5月13日

查看全文
  • Gehad Ismail Sayed,
  • Ghada Dahy,
  • Aboul Ella Hassanien

摘要

Autism spectrum disorder (ASD) children face communication, behavior, and social challenges, making it challenging for caregivers to provide constant supervision and address unusual behaviors. To address this, the proposed system monitors the emotions of an ASD child by analyzing their facial expressions. When the system detects emotions such as fear, sadness, or anger, it generates a spoken avatar of the child's caregiver, to provide immediate support and intervention. The proposed system comprises two main components working concurrently: real-time facial emotion detection and talking avatar generation. For the real-time emotion detection component, an optimized version of the Xception deep-learning architecture is introduced, utilizing a modified Hippopotamus Optimization algorithm (MHO). The optimized Xception model demonstrated superior performance across three benchmark datasets. The model achieved impressive accuracies of 99.61% on Extended CK + 48, 82.68% on FER2013, and 97.18% on autistic children emotions dataset. Additionally, the results demonstrate that MHO algorithm significantly enhances the Xception model's performance and the overall real-time emotion detection system.