Super-Resolusi Menggunakan Metode Vast Receptive Field Berbasis Swin Attention Pada Citra Satelit Resolusi Rendah Untuk Deteksi Kendaraan

Ar Rasyid, Abdul Aziz (2025) Super-Resolusi Menggunakan Metode Vast Receptive Field Berbasis Swin Attention Pada Citra Satelit Resolusi Rendah Untuk Deteksi Kendaraan. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6002231013-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Intelligent Transportation System (ITS) merupakan sistem terintegrasi yang dirancang untuk mengoptimalkan manajemen lalu lintas serta meningkatkan keselamatan pengguna jalan. Salah satu aspek krusial dalam ITS adalah kemampuan memperoleh informasi spasial mengenai kondisi lalu lintas, seperti distribusi kendaraan dan infrastruktur transportasi. Citra satelit menyediakan sumber data yang menjanjikan karena cakupan wilayahnya yang luas, namun keterbatasan resolusi spasialnya dapat memengaruhi akurasi interpretasi visual, khususnya terhadap objek berukuran kecil seperti kendaraan, yang menjadi indikator penting dalam menilai kepadatan lalu lintas dan efisiensi infrastruktur. Penelitian ini memperkenalkan model super-resolusi berbasis kombinasi Vast-Receptive-Field dengan Shifted Window Attention (VapSwinSR) untuk meningkatkan resolusi citra satelit. Kombinasi ini menghasilkan peningkatan ukuran receptive field dengan bobot perhatian yang lebih efisien sehingga menghasilkan kualitas rekonstruksi citra super-resolusi yang lebih baik. Penggunaan citra super-resolusi juga dapat meningkatkan performa deteksi kendaraan. Berdasarkan hasil uji coba performa super-resolusi, model ini menghasilkan peningkatan kualitas PSNR sebesar 2,7% dibandingkan metode berbasis Swin Transformer dengan total parameter yang lebih sedikit, dan meningkat 9,95% dibandingkan metode klasik. Pada pengujian deteksi kendaraan menggunakan YOLOv8, penggunaan citra super-resolusi memberikan peningkatan rata-rata nilai mAP50 sebesar 14,26%. Hasil ini membuktikan bahwa peningkatan resolusi citra satelit tidak hanya memperbaiki kualitas visual, tetapi juga mendukung efektivitas deteksi kendaraan.
============================================================================================================================================
The Intelligent Transportation System (ITS) was an integrated system designed to optimize traffic management and improve road user safety. One crucial aspect of ITS was the ability to obtain spatial information regarding traffic conditions, such as vehicle distribution and transportation infrastructure. Satellite imagery provided a promising data source due to its wide geographic coverage. However, its limited spatial resolution could affect the accuracy of visual interpretation, particularly for small-sized objects like vehicles, which were important indicators for assessing traffic density and infrastructure efficiency. This study introduced a super-resolution model based on the combination of Vast-Receptive-Field and Shifted Window Attention (VapSwinSR) to enhance the resolution of satellite images. This combination increased the receptive field size with more efficient attention weighting, thereby yielding better quality in super-resolved image reconstruction. The use of super-resolved imagery also improved vehicle detection performance. Based on super-resolution performance evaluation, the proposed model achieved a 2.7% improvement in PSNR compared to Swin Transformer-based methods with fewer total parameters, and a 9.95% improvement compared to classical methods. In vehicle detection experiments using YOLOv8, the use of super resolved images led to an average increase of 14.26% in mAP 50 score. These results demonstrated that enhancing satellite image resolution not only improved visual quality but also supported more effective vehicle detection.

Item Type:	Thesis (Masters)
Subjects:	T Technology > T Technology (General) > T385 Visualization--Technique T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Mathematics, Computation, and Data Science > Mathematics > 44101-(S2) Master Thesis
Depositing User:	Abdul Aziz Ar Rasyid
Date Deposited:	04 Aug 2025 07:03
Last Modified:	04 Aug 2025 07:03
URI:	http://repository.its.ac.id/id/eprint/125689

Actions (login required)

View Item