Deteksi Dan Pemantauan Kapal Untuk Maritime Situational Awareness Menggunakan Data SAR Berbasis Kecerdasan Buatan

Oktaviani, Helena Ditya Oktaviani (2026) Deteksi Dan Pemantauan Kapal Untuk Maritime Situational Awareness Menggunakan Data SAR Berbasis Kecerdasan Buatan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5018221007-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Pengawasan kapal di wilayah perairan Indonesia membutuhkan sumber informasi yang mampu mendukung deteksi objek secara non-kooperatif, terutama ketika data identitas kapal dari sistem pelaporan seperti Automatic Identification System (AIS) tidak tersedia atau tidak aktif. Citra Synthetic Aperture Radar (SAR) relevan digunakan karena dapat melakukan pengamatan pada siang maupun malam hari, relatif tidak dipengaruhi tutupan awan, serta tidak bergantung pada sistem kooperatif kapal. Penelitian ini bertujuan membentuk dataset citra patch SAR dari data GFW SAR Vessel Detections, merancang sistem deteksi dan klasifikasi kapal berbasis YOLOv12, serta mengevaluasi kinerja dan kemampuan generalisasi model pada dataset internal dan eksternal. Dataset akhir GFW terdiri atas 534 objek kapal dari 254 scene Sentinel-1 dengan tiga kelas, yaitu Fishing, Cargo, dan Passenger. Pelatihan dilakukan menggunakan lima varian YOLOv12, tiga skenario pembagian dataset, dan dua jumlah epoch, sehingga menghasilkan 30 skenario pelatihan. Hasil validasi menunjukkan bahwa model terbaik adalah YOLOv12N pada split 80:10:10 dengan 50 epoch, dengan mAP50 validasi sebesar 0,79124. Pada pengujian internal GFW, model memperoleh precision 0,59746, recall 0,64069, mAP50 0,75515, dan mAP50-95 0,42588. Namun, pada pengujian eksternal, mAP50 menurun menjadi 0,22419 pada OpenSARShip-1.0 dan 0,21178 pada OpenSARShip-2.0. Hasil tersebut menunjukkan bahwa model dapat mendukung deteksi awal kapal untuk Maritime Situational Awareness pada domain data yang sejenis dengan data latih, tetapi kemampuan generalisasi pada dataset eksternal masih terbatas.
=========================================================================================================================================
Vessel surveillance in Indonesian waters requires information sources capable of supporting non-cooperative object detection, particularly when vessel identity data from reporting systems such as the Automatic Identification System (AIS) are unavailable or inactive. Synthetic Aperture Radar (SAR) imagery is relevant for this purpose because it can acquire observations during both day and night, is relatively unaffected by cloud cover, and does not depend on cooperative vessel-based systems. This study aims to construct a SAR patch image dataset from GFW SAR Vessel Detections, design a YOLOv12-based vessel detection and classification system, and evaluate the model performance and generalization capability on internal and external datasets. The final GFW dataset consists of 534 vessel objects from 254 Sentinel-1 scenes with three classes, namely Fishing, Cargo, and Passenger. Training was conducted using five YOLOv12 variants, three dataset split scenarios, and two epoch settings, resulting in 30 training scenarios. The validation results show that the best model is YOLOv12N with an 80:10:10 split and 50 epochs, achieving a validation mAP50 of 0.79124. In the internal GFW test, the model obtained a precision of 0.59746, recall of 0.64069, mAP50 of 0.75515, and mAP50-95 of 0.42588. However, in the external tests, the mAP50 decreased to 0.22419 on OpenSARShip-1.0 and 0.21178 on OpenSARShip-2.0. These results indicate that the model can support initial vessel detection for Maritime Situational Awareness on data domains similar to the training data, while its generalization capability on external datasets remains limited.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi kapal, Synthetic Aperture Radar, YOLOv12, Deep learning , Maritime Situational Awareness, Vessel detection, Synthetic Aperture Radar, YOLOv12, Deep learning, Maritime Situational Awareness.
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G70.5.I4 Remote sensing Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. V Naval Science > V Naval Science (General)
Divisions:	Faculty of Marine Technology (MARTECH) > Naval Architecture and Shipbuilding Engineering > 36201-(S1) Undergraduate Thesis
Depositing User:	Helena Ditya Oktaviani
Date Deposited:	30 Jul 2026 03:23
Last Modified:	30 Jul 2026 03:23
URI:	http://repository.its.ac.id/id/eprint/140176

Actions (login required)

View Item