Deteksi Hukum Bacaan Tajweed Menggunakan Metode YOLO (You Only Look Once) Berbasis Digital Image Processing

Azizah, Anisa' Nur (2023) Deteksi Hukum Bacaan Tajweed Menggunakan Metode YOLO (You Only Look Once) Berbasis Digital Image Processing. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025211041-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2025. Download (5MB) | Request a copy

Abstract

Ilmu Tajweed merupakan ilmu yang wajib dipelajari bagi seorang muslim yang bertujuan agar dapat membaca Al-Qur'an dengan baik dan benar. Tajweed berisi tentang berbagai macam hukum bacaan dan tata cara pelafalan dalam Al-Qur'an. Pembagian hukum dan aturan yang begitu banyak menyebabkan sedikit orang yang dapat menghafal keseluruhan. Penelitian ini melakukan pendekatan berbasis image processing dalam mendeteksi hukum bacaan Tajweed. Tiap hukum bacaan akan dipelajari melalui fitur-fitur ekstraksi dari suatu scan image mushaf Al-Qur’an. Langkah awal pada penelitian ini adalah pengumpulan data. Selanjutnya melakukan pelabelan kelas sesuai dengan hukum bacaan Tajweed. Dalam penelitian ini terdapat 10 kelas hukum bacaan Tajweed yang akan dilakukan pembelajaran fitur dan pembuatan sistem deteksi menggunakan metode YOLO. Metode YOLO merupakan salah satu metode deteksi objek yang memiliki kinerja baik meskipun dilakukan secara real-time. Metode YOLO memiliki beberapa versi yang terus dikembangkan untuk mengoptimalkan permasalahan deteksi objek. Penelitian ini membandingkan 3 versi terbaru YOLO yaitu YOLO-V5, YOLO-V6, dan YOLO-V7 untuk mendeteksi hukum bacaan Tajweed. Salah satu kendala dalam pendeteksian Tajweed adalah permasalahan small dataset. Penelitian ini melakukan proses augmentasi data untuk mengatasi permasalahan tersebut. Proses augmentasi data mushaf Al-Qur’an dilakukan berdasarkan variasi model warna HSV yaitu hue, saturation, brightness, dan contrast. Berdasarkan hasil uji coba, proses augmentasi dapat meningkatkan performa sistem hingga 65% pada data training dan 47% pada data testing dibandingkan tanpa augmentasi. Perbandingan 3 versi YOLO menyimpulkan bahwa YOLO-V7 lebih baik dari YOLO-V5 dan YOLO-V6 meskipun memiliki waktu training yang paling lama yaitu 12,173 jam. Terlihat pada hasil evaluasi mAP0.5 data testing model YOLO-V5, YOLO-V6, dan YOLO-V7 berturut-turut adalah 71%, 69%, dan 80%, serta tingkat inferensi deteksi berturut-turut adalah 39ms, 36ms, dan 31ms. Hasil ini membuktikan bahwa hasil model penelitian ini cocok untuk deteksi Tajweed dengan baik dan real-time sehingga dapat membantu pemahaman mengenai Ilmu Tajweed.
===================================================================================================================================
Tajweed is a knowledge that must be studied for reading the Al-Qur'an properly and correctly, especially for a Muslim. Tajweed contains various laws of reading and procedures for reciting the Al-Qur'an. The rules of Tajweed are so numerous that not many people can memorize the entire of them. This study used an image processing-based approach for detecting Tajweed. Each Tajweed rule (class) is studied through feature extraction from a scanned image of the Al-Qur'an Mushaf. The initial step in this research is data collection. Next, we do the labeling and annotation according to the Tajweed class. In this research, there are ten Tajweed classes that became learning features and creating a detection system using the YOLO method. The YOLO method is a real-time object detection method that has good performance. The YOLO method has several versions that continue to be developed to optimize object detection problems. This research compares 3 versions of YOLO: YOLO-V5, YOLO-V6, and YOLO-V7 to detect the reading law of Tajweed. One of the challenges in detecting Tajweed is if it has the small dataset. This study proposed the HSV color model such as hue, saturation, brightness, and contrast for data augmentation process to overcome these problems. Based on experimental results, the data augmentation process can improve system performance up to 65% on training dataset and 47% on testing dataset compared to data without augmentation. A comparison of the 3 YOLO versions concluded that YOLO-V7 is better than YOLO-V5 and YOLO-V6 even though it has the longest training time (12.17 hours). It can be seen from the evaluation results of mAP0.5 for the YOLO-V5, YOLO-V6, and YOLO-V7 models are 71%, 69%, and 80% respectively, and the detection inference rate is 39ms, 36ms, and 31ms respectively. These results prove that the results of this study are suitable for detecting Tajweed properly and in real-time, so it can help understanding the learning of Tajweed.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Tajweed, Augmentasi Data, Deteksi Objek, Data Augmentation, Object Detection, YOLO
Subjects:	T Technology > T Technology (General) > T11 Technical writing. Scientific Writing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Anisa' Nur Azizah
Date Deposited:	13 Feb 2023 04:02
Last Modified:	17 Feb 2023 02:19
URI:	http://repository.its.ac.id/id/eprint/96889

Actions (login required)

View Item