Scholarlink

<p>Emotion is a multifaceted response pattern involving neurophysiological and behavioural components. While crucial for social development, affect recognition research has predominantly focused on adults, creating a significant gap in understanding children, particularly those with special needs. Deaf and hard-of-hearing children often face distinct challenges in emotional expression due to barriers in verbal communication. This study presents a novel multimodal database to address this gap, focusing on affect recognition in deaf children through physiological signals. We collected data from 15 deaf participants using the EmotiBit wearable device during video stimuli designed to elicit positive, negative, and neutral emotional responses. The database comprises Electrodermal Activity (EDA), multi-wavelength photoplethysmography (PPG), temperature, and Inertial Measurement Unit (IMU) data. We extracted features from EDA, PPG, and IMU signals for affective classification. To evaluate the dataset, we implemented three distinct multimodal ensemble learning approaches based on a Two-Stage Decision-Level Fusion strategy. These methods—two-level hard voting, stage-wise weighted voting, and hierarchical stacking—utilized SVM, XGBoost, and a Hybrid Neural Network as base classifiers, effectively synergizing their strengths to optimize performance. To our knowledge, this is the first study to collect and analyze a multimodal physiological dataset specifically for affect recognition in deaf children. Our innovative approach and promising findings highlight the potential for developing nonverbal affect recognition systems tailored to this population, ultimately supporting their emotional and social development.</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

Affect Recognition in Deaf Children Using Physiological Signal Measurements

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Kavita Choudhary,
  • Gend Lal Prajapati

摘要

Emotion is a multifaceted response pattern involving neurophysiological and behavioural components. While crucial for social development, affect recognition research has predominantly focused on adults, creating a significant gap in understanding children, particularly those with special needs. Deaf and hard-of-hearing children often face distinct challenges in emotional expression due to barriers in verbal communication. This study presents a novel multimodal database to address this gap, focusing on affect recognition in deaf children through physiological signals. We collected data from 15 deaf participants using the EmotiBit wearable device during video stimuli designed to elicit positive, negative, and neutral emotional responses. The database comprises Electrodermal Activity (EDA), multi-wavelength photoplethysmography (PPG), temperature, and Inertial Measurement Unit (IMU) data. We extracted features from EDA, PPG, and IMU signals for affective classification. To evaluate the dataset, we implemented three distinct multimodal ensemble learning approaches based on a Two-Stage Decision-Level Fusion strategy. These methods—two-level hard voting, stage-wise weighted voting, and hierarchical stacking—utilized SVM, XGBoost, and a Hybrid Neural Network as base classifiers, effectively synergizing their strengths to optimize performance. To our knowledge, this is the first study to collect and analyze a multimodal physiological dataset specifically for affect recognition in deaf children. Our innovative approach and promising findings highlight the potential for developing nonverbal affect recognition systems tailored to this population, ultimately supporting their emotional and social development.