Real-Time Air-Writing Recognition for Arabic Letters Using Deep Learning.

Area of Science:

• Computer Science
• Education Technology
• Cognitive Development

Background:

• Arabic alphabet acquisition is vital for cognitive development but lacks digital tools.
• Existing educational applications for Arabic language learning are insufficient.
• Interactive learning methods can improve memory and retention skills.

Purpose of the Study:

• To develop an interactive web application, SamAbjd, for teaching Arabic letters using air-writing recognition.
• To address the gap in effective digital Arabic language learning resources for children.
• To integrate deep learning techniques with user-centered design for educational purposes.

Main Methods:

• Developed SamAbjd, an interactive web application utilizing deep learning for air-writing recognition.
• Collected and preprocessed a dataset of 31,349 annotated Arabic letter images.
• Experimented with Convolutional Neural Network (CNN) and VGG16 models for character recognition.

Main Results:

• A seven-layer CNN model without dropout achieved 96.40% testing accuracy.
• The best-performing model demonstrated high precision (96.44%) and F1-score (96.43%).
• The SamAbjd application provides a user-friendly interface for learning Arabic letters.

Conclusions:

• SamAbjd effectively teaches Arabic letters using deep learning and air-writing recognition.
• The application successfully meets educational needs by incorporating user feedback and advanced technology.
• This tool enhances cognitive development by improving memory and retention in Arab children.