Optimizing Tajwid Rule Detection in Quranic Verses Using Deep Learning and Transfer Learning Techniques

Abstrak

This study addresses the difficulty learners face in mastering tajwid rules especially nun sakinah and tanwin where traditional instruction often leads to recitation errors. The research aimed to develop and evaluate an automated tajwid detection system using the YOLOv8 deep learning model and compare its performance with YOLOv5. The dataset consisted of 125 Qur’anic page images manually annotated into 600 instances across five tajwid rule classes. Preprocessing included image resizing, augmentation, and dataset splitting into training, validation, and testing subsets. Transfer learning was applied during model training, and performance was measured using precision, recall, and mean Average Precision. A web-based system built with the Flask framework was developed to support both image uploads and real-time camera detection. The results show that YOLOv8 outperformed YOLOv5, achieving 94.6% accuracy, 96.5% precision, and 90.2% recall. Class performance varied, with the highest accuracy for Idgham Bilaghunnah (98.7%) and the lowest for Iqlab (91.1%) due to dataset imbalance. The findings indicate that the Smart Tajwid system effectively integrates computer vision with Qur’anic education, offering a practical tool for learners. Future work should focus on expanding the dataset and optimizing the model to enhance detection consistency across all tajwid categories.

Identifier

• OpenAlex: https://openalex.org/W4417528463
• DOI: https://doi.org/10.38035/jemsi.v7i2.7063