RIVER AREA SEGMENTATION USING SENTINEL-1 SAR IMAGERY WITH DEEP LEARNING APPROACH

Dewi, Ni Putu Karisma (2025) RIVER AREA SEGMENTATION USING SENTINEL-1 SAR IMAGERY WITH DEEP LEARNING APPROACH. Undergraduate thesis, Universitas Pendidikan Ganesha.

[img] Text (COVER)
2115101059 - COVER.pdf
Download (570kB)
[img] Text (ABSTRAK)
2115101059 - ABSTRAK.pdf
Download (236kB)
[img] Text (BAB 1 PENDAHULUAN)
2115101059 - BAB 1 PENDAHULUAN.pdf
Download (250kB)
[img] Text (BAB 2 KAJIAN TEORI)
2115101059 - BAB 2 KAJIAN TEORI.pdf
Restricted to Repository staff only
Download (1MB) | Request a copy
[img] Text (BAB 3 METODELOGI PENELITIAN)
2115101059 - BAB 3 METODELOGI PENELITIAN.pdf
Restricted to Repository staff only
Download (596kB) | Request a copy
[img] Text (BAB 4 HASIL DAN PEMBAHASAN)
2115101059 - BAB 4 HASIL DAN PEMBAHASAN.pdf
Restricted to Repository staff only
Download (10MB) | Request a copy
[img] Text (BAB 5 PENUTUP)
2115101059 - BAB 5 PENUTUP.pdf
Restricted to Repository staff only
Download (251kB) | Request a copy
[img] Text (DAFTAR PUSTAKA)
2115101059 - DAFTAR PUSTAKA.pdf
Download (256kB)
[img] Text (LAMPIRAN)
2115101059 - LAMPIRAN.pdf
Download (186kB)

Abstract

This research focuses on creating a river dataset using satellite data and performing semantic river segmentation with deep learning architectures. The motivation behind this study is the potential of satellites to enable more efficient global river mapping, while deep learning architectures offer a more effective approach to segment areas that specifically represent rivers. The dataset was constructed using Sentinel-1 C-Band SAR Ground Range Detected Interferometric Wide Swath imagery, obtained through the Google Earth Engine platform. The satellite data was labelled with binary labels to distinguish between river and non-river areas. The dataset preparation process was carried out using Python in Jupyter Notebook. The generated dataset was then used for modelling, where two deep learning algorithms, U-Net and DeepLabv3+, were trained on the same dataset. The TensorFlow and Keras libraries were utilized to develop and train the models. The results showed promising performance, with U-Net achieving a Dice Coefficient score of 96% and DeepLabv3+ scoring 95%. During testing, both models demonstrated strong performance. U-Net achieved a Recall of 95% and Precision of 91%, while DeepLabv3+ achieved a Recall of 90% and Precision of 89%. From the analysis, it was observed that U-Net tends to produce detailed segmentations, where the predicted images closely resemble the labels. In contrast, DeepLabv3+ segments using broader, more global features. Based on these findings, this research highlights the potential for further development to address real-world applications, such as flood monitoring, drought mitigation, and water flow management. Future studies could focus on expanding the dataset and exploring more advanced architectures to enhance performance and applicability. Keyword: Semantic Segmentation, Sentinel-1 C-Band SAR Ground Range Detected Interferometric Wide Swath, U-Net, DeepLabv3+, River.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: Semantic Segmentation, Sentinel-1 C-Band SAR Ground Range Detected Interferometric Wide Swath, U-Net, DeepLabv3+, River.
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
Divisions: Fakultas Teknik dan Kejuruan > Jurusan Teknik Informatika > Program Studi Ilmu Komputer (S1)
Depositing User: Ni Putu Karisma Dewi
Date Deposited: 30 Jan 2025 04:30
Last Modified: 30 Jan 2025 04:30
URI: http://repo.undiksha.ac.id/id/eprint/22935

Actions (login required)

View Item View Item