Pengembangan Metode Deep Learning untuk Identifikasi Kapal di Perairan Maluku Utara Menggunakan Data AIS dan Citra SAR Sentinel-1

Insani, Hanif Rafdhiansyah (2026) Pengembangan Metode Deep Learning untuk Identifikasi Kapal di Perairan Maluku Utara Menggunakan Data AIS dan Citra SAR Sentinel-1. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6016242001-Master_thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Keterbatasan sistem pemantauan berbasis Automatic Identification System (AIS) yang bergantung pada transmisi aktif kapal mendorong pemanfaatan citra Synthetic Aperture Radar (SAR) Sentinel-1 sebagai alternatif pemantauan maritim yang independen terhadap kondisi cuaca maupun siklus siang dan malam. Penelitian ini mengembangkan dan membandingkan kinerja model deteksi kapal berbasis deep learning YOLOv11s dan YOLOv11m pada citra SAR Sentinel-1 di perairan Maluku Utara dengan mengintegrasikan data AIS dari Global Fishing Watch sebagai referensi spasial. Pra-pemrosesan citra meliputi koreksi orbit, thermal noise removal, kalibrasi radiometrik (σ⁰), speckle filtering, dan terrain correction, menghasilkan komposit RGB (R=VV, G=VH, B=VV+VH) yang dipotong menjadi 360 tile berukuran 640×640 piksel dengan 98 objek kapal teranotasi. Hasil evaluasi menunjukkan YOLOv11s unggul dengan mAP@0,5 sebesar 0,759, precision 1,00, recall 0,91, dan F1-score 0,70, melampaui ambang batas kelayakan minimum. Sebaliknya, YOLOv11m hanya mencapai mAP@0,5 sebesar 0,477 akibat early stopping pada epoch ke-40 yang disebabkan ketidaksesuaian kapasitas arsitektur dengan ukuran dataset yang terbatas. Analisis kesesuaian spasial dari 159 titik deteksi bersih menunjukkan 18 deteksi (11,3%) terverifikasi dalam buffer ≤10 km, 33 deteksi (20,8%) dalam buffer 10–20 km, dan 108 deteksi (67,9%) dikategorikan sebagai dark vessel. Rasio AIS terhadap total deteksi YOLO yang hanya 34,0% mengonfirmasi bahwa pemantauan berbasis SAR mampu mengidentifikasi jauh lebih banyak kapal dibandingkan sistem AIS konvensional, menegaskan potensi integrasi SAR dan AIS sebagai pendekatan komplementer dalam sistem pemantauan maritim di perairan Indonesia Timur. ======================================================================================================================================
The limitations of Automatic Identification System (AIS)-based monitoring, which relies on active vessel transmission, have driven the utilization of Synthetic Aperture Radar (SAR) Sentinel-1 imagery as an alternative maritime monitoring approach that operates independently of weather conditions and day-night cycles. This study develops and compares the performance of YOLOv11s and YOLOv11m deep learning-based vessel detection models on Sentinel-1 SAR imagery over North Maluku waters, integrating AIS data from Global Fishing Watch as spatial reference. Image preprocessing encompassed orbit correction, thermal noise removal, radiometric calibration (σ⁰), speckle filtering, and terrain correction, yielding RGB composites (R=VV, G=VH, B=VV+VH) subsequently tiled into 360 patches of 640×640 pixels containing 98 annotated vessel objects. Evaluation results demonstrated that YOLOv11s outperformed YOLOv11m, achieving an mAP@0.5 of 0.759, maximum precision of 1.00, maximum recall of 0.91, and maximum F1-score of 0.70, surpassing the established minimum performance threshold. In contrast, YOLOv11m attained only an mAP@0.5 of 0.477 due to early stopping at epoch 40, attributed to the mismatch between the model's architectural capacity and the limited dataset size. Spatial correspondence analysis of 159 clean detections revealed that 18 detections (11.3%) were verified within a ≤10 km buffer, 33 detections (20.8%) within a 10–20 km buffer, and 108 detections (67.9%) were categorized as dark vessels with no corresponding AIS records. The AIS-to-YOLO detection ratio of only 34.0% confirms that SAR-based monitoring identifies substantially more vessels than conventional AIS systems, underscoring the potential of SAR-AIS integration as a complementary approach for comprehensive maritime surveillance in eastern Indonesian waters.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	AIS, Deep Learning, Deteksi Kapal, Perairan Maluku Utara, SAR, Sentinel-1, YOLO, AIS, Deep Learning, SAR, Sea of North Maluku, Sentinel-1, Vessel Detection, YOLO
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G70.5.I4 Remote sensing
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29101-(S2) Master Thesis
Depositing User:	Hanif Rafdhiansyah Insani
Date Deposited:	23 Jul 2026 02:12
Last Modified:	23 Jul 2026 02:12
URI:	http://repository.its.ac.id/id/eprint/136317

Actions (login required)

View Item