BathySAR-Net V.2: Model Berbasis Convolutional Neural Network Untuk Memprediksi Kedalaman Zona Pesisir Menggunakan Citra Synthetic Aperture Radar (SAR) Sentinel-1

Putri, Belvana Eka (2026) BathySAR-Net V.2: Model Berbasis Convolutional Neural Network Untuk Memprediksi Kedalaman Zona Pesisir Menggunakan Citra Synthetic Aperture Radar (SAR) Sentinel-1. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6016251018-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Pemetaan batimetri perairan dangkal penting untuk mendukung keselamatan navigasi dan pengelolaan wilayah pesisir, namun survei batimetri konvensional masih memiliki keterbatasan biaya, waktu, dan cakupan. Penelitian ini bertujuan mengembangkan BathySAR-Net V.2, model berbasis Convolutional Neural Network (CNN) untuk memprediksi kedalaman perairan menggunakan citra Sentinel-1 SAR di wilayah pesisir Kota Pekalongan. Model memanfaatkan arsitektur DenseNet, Gated Feature-wise Linear Modulation (Gated FiLM), serta pre-training menggunakan data BATNAS. Evaluasi dilakukan menggunakan metrik Root Mean Square Error (RMSE) dan Mean Absolute Percentage Error (MAPE) terhadap data survei lapangan. Model terbaik menghasilkan RMSE sebesar 0,7704 m dan akurasi prediksi sebesar 92,13% pada data uji yang tidak digunakan selama pelatihan. Dibandingkan BathySAR-Net V.1 berbasis Artificial Neural Network (ANN) dengan menggunakan data yang sama, model yang diusulkan berhasil meningkatkan akurasi dengan menurunkan RMSE dari 0,9381 m menjadi 0,7704 m serta meningkatkan akurasi dari 84,94% menjadi 92,13%. Hasil penelitian menunjukkan bahwa pendekatan CNN efektif untuk meningkatkan akurasi estimasi batimetri berbasis citra Sentinel-1 SAR.
======================================================================================================================================
Bathymetric mapping in shallow coastal waters is essential for navigation safety and coastal management. However, conventional acoustic surveys are often limited by high costs, long acquisition times, and restricted spatial coverage. This study proposes BathySAR-Net V.2, a Convolutional Neural Network (CNN)-based model for water depth estimation using Sentinel-1 Synthetic Aperture Radar (SAR) imagery in the coastal waters of Pekalongan, Indonesia. The proposed model integrates a DenseNet architecture, Gated Feature-wise Linear Modulation (Gated FiLM), and BATNAS-based pre-training to improve prediction performance. Model accuracy was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) against in-situ bathymetric survey data. The best model achieved an RMSE of 0.7704 m and an Accuracy of 92.13% on an unseen Sentinel-1 dataset, demonstrating good generalization capability. Compared with the previous BathySAR-Net V.1 based on an Artificial Neural Network (ANN) using the same data, the proposed model improved prediction accuracy by reducing the RMSE from 0.9381 m to 0.7704 m and the Accuracy from 84.94% to 92.13%. These results demonstrate the effectiveness of the CNN-based approach for SAR-derived bathymetric mapping.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Satellite-Derived Bathymetry, Synthetic Aperture Radar, Sentinel-1, Convolutional Neural Network, Batimetri. ======================================================================================================================== Satellite-Derived Bathymetry, Synthetic Aperture Radar, Sentinel-1, Convolutional Neural Network, Bathymetry
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G70.217 Geospatial data T Technology > TC Hydraulic engineering. Ocean engineering > TC203.5 Coastal engineering
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29101-(S2) Master Thesis
Depositing User:	Belvana Eka Putri
Date Deposited:	30 Jul 2026 06:41
Last Modified:	30 Jul 2026 06:41
URI:	http://repository.its.ac.id/id/eprint/139744

Actions (login required)

View Item