Identifikasi Hukum Tajwid pada Citra Teks Al Quran menggunakan SSD MobileNet v2

Abstract

Tajweed contains a set of rules for reciting the Qur'an correctly. These rules must be complied with to ensure each letter is pronounced accurately. Arabic script and language compose the Qur'an, yet not all readers are fluent in Arabic. Tajweed serves as a guide to prevent readers from making mistakes when reciting the Qur'an that could alter the meaning. However, Tajweed rules are quite numerous and diverse, causing readers to struggle in memorizing these rules. To address this issue, a preliminary development of a Quran reading assistance system will be established, focusing on detecting Tajweed rules in images of Quranic text. SSD MobileNet v2, a Deep Learning technique for object detection, will be utilized for detecting Tajweed rules. The development of the Tajweed rule identification model begins with the data collection stage by capturing screens of the Al-Quran text pages from the Kemenag Qur'an Application. A total of 520 collected data were divided into 80:10:10 for training, validation, and test data, respectively. All data were subsequently annotated and enclosed in bounding boxes using the tool labelImg. The pre-trained model, SSD MobileNet V2 FPNLite 320x320, was used as the initial weight configuration of the model. Then the identification model was constructed during the training stage using training and validation data. The reliability of the constructed model was tested using test data. The test results indicated that the model could successfully recognize two Tajwid rules, Mad Aridlisukun and Mad Layyin, achieving the minimum loss around 0.15 and the maximum precision around 0.96.