Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Hacettepe Üniversitesi, Fen Bilimleri Enstitüsü, Geomatik Mühendisliği A.B.D., Türkiye
Tezin Onay Tarihi: 2025
Tezin Dili: İngilizce
Öğrenci: MELİH ALTAY
Danışman: Mustafa Türker
Özet:
This study presents a deep learning-based framework for the semantic segmentation of burned forest areas using exclusively post-fire PlanetScope imagery, which offers high spatial resolution (3 meters) and daily temporal coverage. A key novelty of the research lies in its application of three advanced convolutional neural network architectures—U-Net, DeepLabV3+, and Fully Convolutional Network (FCN)—within a transfer learning framework to enable effective adaptation to diverse ecological conditions. The methodology was validated across three geographically and ecologically distinct fire-affected regions in Türkiye: Manavgat (MG), Bodrum (BD), and Çanakkale (CK), each characterized by different vegetation types, land cover patterns, and fire intensities. Experimental results showed that DeepLabV3+ achieved the best overall performance in BD and CK regions, with F1-Scrs of 79% and 85%, respectively. U-Net outperformed others in the MG region with a 94% F1-Scr and 95% prc for burned areas. FCN showed high rc for burned areas in MG (96%) but struggled in distinguishing unburned zones, particularly in BD and CK. The models were trained and evaluated using over 4,100 image patches, with transfer learning enabling consistent generalization across regions.
The
study’s originality stems from (i) the exclusive use of post-fire
high-resolution PlanetScope imagery, (ii) the comparative evaluation of three
semantic segmentation models across ecologically diverse fire regions, and
(iii) the incorporation of transfer learning to enhance model generalizability
in data-scarce post-disaster scenarios. Findings highlight the potential of
DeepLabV3+ as a robust and transferable model for operational burned area
detection, supporting timely post-fire damage assessment and ecological
monitoring in Mediterranean forest ecosystems.