Scholarlink

<p>Sign language (SL) is a significant communication method for individuals with hearing impairments, using hand gestures to convey letters, words, and sentences. However, several people are unfamiliar with SL, creating a communication gap. An intelligent SL recognition framework must bridge this gap by accurately interpreting hand gestures into text or speech. The system uses deep learning (DL) to ensure real-time, efficient, and inclusive communication between hearing-impaired individuals and others. Hearing aid devices are notable technological developments that assist hearing-impaired individuals in communicating with others. It also helps individuals with partial hearing loss, while those with complete hearing loss rely entirely on SL for communication. Advancements in machine learning (ML) and computer vision (CV) have led to effective recognition and interpretation of SL gestures. This paper proposes an Optimized Fusion-Based Transfer Learning for Sign Language Recognition Using the Bald Eagle Search Algorithm (OFTLSLR-BESA) model. The main aim of the proposed OFTLSLR-BESA technique relies on improving the recognition system of SL to assist individuals with hearing and speech impairments. The adaptive bilateral filtering (ABF) technique initially performs the image pre-processing step to eliminate the unwd noise from input image data. Furthermore, the feature extraction process employs a fusion of InceptionV3, MobileNetV2, NASNetMobile and ResNet50V2 models. Moreover, the random vector functional link (RVFL) method is implemented for classification. Finally, the RVFL model’s hyperparameter tuning is performed using the Bald Eagle Search (BES) method. The OFTLSLR-BESA approach is evaluated under the American SL alphabet dataset. The experimental analysis of the OFTLSLR-BESA approach portrayed a superior accuracy value of 99.64% over existing models.</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

Deep fusion based transfer learning with bald eagle search algorithm for sign language recognition to assist individuals with hearing and speech impairments

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月24日

查看全文
  • Manar Ahmed Hamza,
  • Alanoud Subahi,
  • Nouf Atiahallah Alghanmi,
  • Nadhem Nemri

摘要

Sign language (SL) is a significant communication method for individuals with hearing impairments, using hand gestures to convey letters, words, and sentences. However, several people are unfamiliar with SL, creating a communication gap. An intelligent SL recognition framework must bridge this gap by accurately interpreting hand gestures into text or speech. The system uses deep learning (DL) to ensure real-time, efficient, and inclusive communication between hearing-impaired individuals and others. Hearing aid devices are notable technological developments that assist hearing-impaired individuals in communicating with others. It also helps individuals with partial hearing loss, while those with complete hearing loss rely entirely on SL for communication. Advancements in machine learning (ML) and computer vision (CV) have led to effective recognition and interpretation of SL gestures. This paper proposes an Optimized Fusion-Based Transfer Learning for Sign Language Recognition Using the Bald Eagle Search Algorithm (OFTLSLR-BESA) model. The main aim of the proposed OFTLSLR-BESA technique relies on improving the recognition system of SL to assist individuals with hearing and speech impairments. The adaptive bilateral filtering (ABF) technique initially performs the image pre-processing step to eliminate the unwd noise from input image data. Furthermore, the feature extraction process employs a fusion of InceptionV3, MobileNetV2, NASNetMobile and ResNet50V2 models. Moreover, the random vector functional link (RVFL) method is implemented for classification. Finally, the RVFL model’s hyperparameter tuning is performed using the Bald Eagle Search (BES) method. The OFTLSLR-BESA approach is evaluated under the American SL alphabet dataset. The experimental analysis of the OFTLSLR-BESA approach portrayed a superior accuracy value of 99.64% over existing models.