Pengenalan Dan Transliterasi Aksara Sunda Menggunakan YOLO Dan LSTM

Nawaf, Mohammad Kamal (2025) Pengenalan Dan Transliterasi Aksara Sunda Menggunakan YOLO Dan LSTM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211180-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (24MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk membangun model yang mampu mendeteksi dan melakukan transliterasi aksara Sunda. Deteksi karakter aksara Sunda diperlukan karena aksara ini merupakan simbol identitas masyarakat Sunda yang memiliki keunikan dan kompleksitas tersendiri. Upaya ini juga mendukung pelestarian budaya melalui teknologi modern.
Penelitian memanfaatkan model YOLO untuk mendeteksi aksara Sunda dengan dataset objek deteksi. Dataset terdiri dari karakter aksara Sunda dalam citra sintetik yang melalui proses pelabelan dan anotasi serta praproses dan augmentation sebelum digunakan dalam pelatihan model. Proses transliterasi dilakukan dengan menggunakan model sequence-to-sequence yang menggunakan Long Short-Term Memory (LSTM) untuk melakukan transliterasi aksara Sunda menjadi teks Latin. Hasil dari deteksi karakter digunakan sebagai masukan untuk transliterasi, hasil deteksi dilakukan praproses terlebih dahulu yaitu pengurutan berdasarkan koordinat x bounding box, ekstraksi nama kelas, dan parsing sesuai dengan format masukkan model transliterasi. Hasil dari praproses dimasukkan ke pada model transliterasi, kemudian model transliterasi mengeluarkan teks yang berupa hasil transliterasi dari aksara Sunda ke teks Latin.
Hasil dari penelitian ini menghasilkan model objek deteksi dengan performa terbaik yang dilatih menggunakan YOLOv12 dan dataset augmentation dengan metrik evaluasi recall sebanyak 93,4% , precision pada 94,8% , mAP@50 sebanyak 96,7%, dan mAP@50-95 pada 90,4%. Penelitian ini juga menghasilkan model transliterasi dengan performa terbaik menggunakan BiLSTM menggunakan dataset dengan 10.000 baris dengan metrik evaluasi CER sebanyak 1,30% dan WER sebanyak 1,35%.
============================================================================================================================
This study aims to build a model capable of detecting and perform transliteration on Sundanese script. Detection of Sundanese script is needed because this script is a symbol of Sundanese identity that has it's own uniqueness and complexity. This study also attempts to preserved culture through modern technology.
This study uses the YOLO model to detect Sundanese script, trained on an object detection dataset. The dataset contains Sundanese characters embedded in synthetically generated images, which are then processed through labeling and annotation, preprocessing, and augmentation before being used for training. Transliteration is performed using a sequence-to-sequence model based on LSTM to convert Sundanese script into Latin script. The output of the detected characters is then used as input for transliteration, where it first undergoes preprocessing steps namely sorting by the x coordinate of the bounding box, extraction of the name of the class, and parsing by following the format for the transliteration model. The result of the preprocessing is then inserted into the transliteration model, which the transliteration model's output is the transliterated Latin text from Sundanese script.
The results of this study produced object detection models, with the best performance of a model trained using YOLOv12 and augmented dataset, achieving evaluation metrics of 93,4% recall, 94,8% precision, 96,7% mAP@50, and 90,4% mAP@50–95. This study also produced transliteration models, with the best performing model using BiLSTM trained on a dataset with 10.000 rows, achieving Character Error Rate (CER) of 1,30% and Word Error Rate (WER) of 1,35%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Aksara Sunda, Deteksi Karakter, Transliterasi, YOLO, LSTM,Sundanese Script, Character Detection, Transliteration
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Mohammad Kamal Nawaf
Date Deposited:	31 Jul 2025 02:30
Last Modified:	31 Jul 2025 02:30
URI:	http://repository.its.ac.id/id/eprint/123821

Actions (login required)

View Item