Optimizing Sign Language Recognition Through a Tailored MobileNet Self-Attention Framework
Year: 2025
Type: article
Abstract: The growing demand for accessible communication technology for the deaf and hearing-impaired illustrates the importance of systems for recognizing sign language that are both accurate and deployable on resource-constrained platforms. Many existing deep learning solutions deliver strong recognition performance but rely on computationally intensive architectures, making them unsuitable for real-time use in mobile and embedded environments. This research introduces a lightweight framework that combines Tailored MobileNet with a Self-Attention module for discriminative feature extraction and integrates optimized machine learning classifiers for efficient prediction. Systematic hyperparameter optimization using Grid Search and cross-validation showed the k-Nearest Neighbors classifier as the most successful classifier. The proposed model was evaluated on four image-based datasets and on one video dataset demonstrating the robustness and effectiveness of the approach, with 99% accuracy across multiple datasets. By achieving high recognition accuracy, model compactness, and computational efficiency on benchmark datasets, this research provides a foundation for the development of practical, scalable sign language recognition systems, highlighting a promising direction for future work in mobile and embedded assistive technologies.
Source: Applied Sciences
DOI: 10.3390/app152312622
Authors: Irfan Qutab, Laura Po, Federica Rollo, Wahab Naqvi
Language: English
Subfield: Human-Computer Interaction
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W4416850927