Akademik Veri Yönetim Sistemi
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025 (SCI-Expanded)
Natural disasters, especially earthquakes, require rapid and accurate damage assessment for effective response and recovery strategies. In this paper, TE23D (Turkey Earthquakes of 6 February 2023 Dataset) dataset consisting of 1183 images and 2080 polygons labelled as damaged was developed using the satellite images covering 10 cities and ∼481,72km2 taken after the earthquakes that occurred on 6 February 2023 in Turkey, and the dataset was evaluated for benchmark results using various deep learning-based object detection techniques by designing and implementing a system to detect damaged areas. Obtaining pre-earthquake images is widely recognized in the literature as beneficial for enhancing damage detection, as it allows for a more accurate assessment of changes caused by disasters. However, in many cases, including ours, it is not feasible to access recent pre-disaster images quickly enough for immediate damage analysis. Therefore, we focused solely on using postearthquake images to identify and label damaged areas, without pre-event imagery. The dataset is designed to perform damage assessment by emphasising the labelling of areas directly affected by post-disaster imagery. To evaluate the effectiveness of this approach, we trained state-of-the-art segmentation models on the dataset, including BEiT, DPT, Mask R-CNN, MobileViT, U-Net, U-Net++, and SegFormer. Among these, the SegFormer model achieved the best performance, with 91,92% Overall Pixel Accuracy (OPA) and 74,45% Intersection over Union for the damaged class (IoUD), demonstrating that labeling damaged areas directly on post-event imagery can yield effective results for damage detection. The findings emphasize the crucial role of high-quality, targeted datasets like TE23D in accelerating disaster response efforts, particularly for earthquake-related damage. By offering a focused benchmark, this dataset enables an efficient and accurate identification of areas most severely affected by earthquakes. This capability for rapid damage assessment is essential for prioritizing emergency response efforts and directing aid to the most critical locations, ultimately helping to save lives and expedite the recovery process. While TE23D is tailored to the context of the February 2023 Turkey earthquakes, its methodology can serve as a model for similar disaster scenarios, highlighting the importance of well-curated datasets in improving the effectiveness of damage detection models across different contexts.