Identifikasi Pipa Bawah Laut Dengan Data Side Scan Sonar Berbasis Model YOLO (You Only Look Once)

Wibowo, Septian Adhe (2026) Identifikasi Pipa Bawah Laut Dengan Data Side Scan Sonar Berbasis Model YOLO (You Only Look Once). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016221107-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Pipa bawah laut merupakan infrastruktur vital dalam sistem transportasi minyak dan gas lepas pantai yang memerlukan pemantauan kondisi secara berkala. Metode inspeksi konvensional berbasis deteksi manual dinilai tidak efisien karena membutuhkan waktu lama, biaya tinggi, dan ketergantungan pada interpretasi manual operator. Penelitian ini mengembangkan metode identifikasi pipa bawah laut secara otomatis menggunakan data Side Scan Sonar (SSS) yang diintegrasikan dengan model deep learning YOLO11m-seg. Tahapan penelitian meliputi pengolahan data SSS menggunakan SonarWiz 8 dengan koreksi Empirical Gain Normalization, nadir Filter, dan destripe Filter pada resolusi 0,250 meter per piksel, pembuatan dataset 1.000 tile citra berukuran 640×640 piksel dengan anotasi polyline mask menggunakan platform Roboflow, pelatihan model dengan pendekatan transfer learning berbasis bobot pralatih COCO pada library Google Colab, serta transformasi koordinat piksel ke sistem koordinat menggunakan transformasi affine. Hasil pelatihan menunjukkan konvergensi yang stabil dengan model terbaik pada epoch ke-54 tanpa indikasi overfitting. Uji presisi pada data testing menghasilkan nilai Box Precision 1,0000, Recall 0,9985, mAP 0.50 0,9950, mAP 0.50 -95 0,9167, serta Mask Precision 0,9900, Recall 0,9885, mAP 0.50 0,9865, dan mAP 0.50 -95 0,6350. Pengujian pada data dari proyek survei berbeda menghasilkan detection rate 84% dengan rata-rata confidence 0,8148, mengindikasikan kemampuan generalisasi lintas-domain yang bermakna. Uji akurasi koordinat menggunakan RMSE menghasilkan nilai total 2D sebesar 0,206 meter, setara dengan 0,824 piksel, yang mengkategorikan akurasi posisi hasil deteksi ke dalam sub-pixel accuracy. Penelitian ini membuktikan bahwa integrasi model YOLO11m-seg dengan transformasi affine mampu menghasilkan identifikasi pipa bawah laut secara otomatis dengan presisi dan akurasi posisi yang tinggi berbasis data SSS.
======================================================================================================================================
Subsea pipelines are critical infrastructures in offshore oil and gas transportation systems that require regular condition monitoring. Conventional inspection methods based on manual detection are considered inefficient due to their high time consumption, operational costs, and dependence on operator interpretation. This study develops an automated subsea pipeline identification method using Side Scan Sonar (SSS) data integrated with the YOLO11m-seg deep learning model. The research workflow consists of SSS data processing in SonarWiz 8 through Empirical Gain Normalization, Nadir Filter, and Destripe Filter corrections at a spatial resolution of 0.250 meters per pixel; preparation of a dataset comprising 1,000 image tiles with dimensions of 640 × 640 pixels annotated using polyline masks in the Roboflow platform; model training using a transfer learning approach based on COCO pre-trained weights in Google Colab; and transformation of pixel coordinates into spatial coordinates using an affine transformation. The training results demonstrated stable convergence, with the best-performing model obtained at epoch 54 without indications of overfitting. Evaluation on the testing dataset yielded Box Precision, Recall, mAP 0.50 , and mAP 0.50 -95 values of 1.0000, 0.9985, 0.9950, and 0.9167, respectively, while Mask Precision, Recall, mAP 0.50 , and mAP 0.50 -95 reached 0.9900, 0.9885, 0.9865, and 0.6350, respectively. Testing on data acquired from a different survey project achieved a detection rate of 84% with an average confidence score of 0.8148, indicating meaningful cross-domain generalization capability. Positional accuracy assessment using Root Mean Square Error (RMSE) resulted in a total two-dimensional RMSE of 0.206 meters, equivalent to 0.824 pixels, thereby categorizing the detection results as achieving subpixel accuracy. The findings demonstrate that the integration of the YOLO11m-seg model with affine coordinate transformation enables automatic subsea pipeline identification with high detection precision and positional accuracy based on Side Scan Sonar data.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pipa bawah laut, YOLO, Deep Learning, Deteksi Objek. Side Scan Sonar, Submarine Pipeline, Object Detection.
Subjects:	Q Science Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Septian Adhe Wibowo
Date Deposited:	21 Jul 2026 01:12
Last Modified:	21 Jul 2026 01:12
URI:	http://repository.its.ac.id/id/eprint/135902

Actions (login required)

View Item