Scholarlink

<p>People with verbal communication challenges often use sign language, but many find it difficult to understand their messages. Researchers have worked together to create a system that helps people who cannot speak communicate with those who use sign language. Even with some flaws, the system achieved its goal successfully. Our main goal in this research is to create Sign Language Recognition (SLR) technology that goes beyond the limitations of the past. This study presents the Advanced Hybrid Convolutional Neural Network (AdH-CNN), which is an ensemble model that merges the VGG16 and ResNet50 neural networks. Combining VGG16’s detail capture with ResNet50’s deep representation improves feature variety, resulting in better recognition. This combination improves accuracy on the datasets “BdSL_OPSA22_STATIC1” and “BdSL_OPSA22_STATIC2”. Moreover, the simplicity of VGG16 balances the complexity of ResNet50, resulting in a strong but manageable computational need. The datasets used in this study are our own and include backgrounds with unclear features. Each dataset contains 24,615 images of Bangla characters and numbers. Finally, the “AdH-CNN” model outperforms earlier models, achieving an accuracy of 96.51 percent on “BdSL_OPSA22_STATIC1” and 97.24 percent on “BdSL_OPSA22_STATIC2”.</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

Computer Vision-based Advanced Hybrid Convolutional Neural Network to Recognize Sign Language for both Numeral and Alphabet Signs

  • 期刊论文

  • 发表日期：

    2025年7月11日

查看全文
  • Kabiratun Ummi Oyshe,
  • Md Ikramul Haque Nirjon,
  • Muhammad Aminur Rahaman,
  • Tanoy Debnath,
  • Utpol Kanti Das,
  • Anichur Rahman

摘要

People with verbal communication challenges often use sign language, but many find it difficult to understand their messages. Researchers have worked together to create a system that helps people who cannot speak communicate with those who use sign language. Even with some flaws, the system achieved its goal successfully. Our main goal in this research is to create Sign Language Recognition (SLR) technology that goes beyond the limitations of the past. This study presents the Advanced Hybrid Convolutional Neural Network (AdH-CNN), which is an ensemble model that merges the VGG16 and ResNet50 neural networks. Combining VGG16’s detail capture with ResNet50’s deep representation improves feature variety, resulting in better recognition. This combination improves accuracy on the datasets “BdSL_OPSA22_STATIC1” and “BdSL_OPSA22_STATIC2”. Moreover, the simplicity of VGG16 balances the complexity of ResNet50, resulting in a strong but manageable computational need. The datasets used in this study are our own and include backgrounds with unclear features. Each dataset contains 24,615 images of Bangla characters and numbers. Finally, the “AdH-CNN” model outperforms earlier models, achieving an accuracy of 96.51 percent on “BdSL_OPSA22_STATIC1” and 97.24 percent on “BdSL_OPSA22_STATIC2”.