QuickQuakeBuildings: Post-earthquake SAR-Optical Dataset for Quick Damaged-building Detection

TL;DR

Post-earthquake damaged-building detection is hampered by the lack of labeled data. The authors present QuickQuakeBuildings, a dataset combining post-event SAR and optical imagery with building footprints for Islahiye, Turkey, following the 2023 Kahramanmaras earthquakes, containing thousands of buildings and four patches per building. They formulate damage detection as a binary image-classification task and provide baselines across SAR, optical, and fused modalities, illustrating that optical data generally outperform SAR while multimodal fusion offers additional gains. This dataset and its benchmarks enable rapid development of robust post-disaster assessment methods and demonstrate the value of multimodal data for fast response in future earthquakes.

Abstract

Quick and automated earthquake-damaged building detection from post-event satellite imagery is crucial, yet it is challenging due to the scarcity of training data required to develop robust algorithms. This letter presents the first dataset dedicated to detecting earthquake-damaged buildings from post-event very high resolution (VHR) Synthetic Aperture Radar (SAR) and optical imagery. Utilizing open satellite imagery and annotations acquired after the 2023 Turkey-Syria earthquakes, we deliver a dataset of coregistered building footprints and satellite image patches of both SAR and optical data, encompassing more than four thousand buildings. The task of damaged building detection is formulated as a binary image classification problem, that can also be treated as an anomaly detection problem due to extreme class imbalance. We provide baseline methods and results to serve as references for comparison. Researchers can utilize this dataset to expedite algorithm development, facilitating the rapid detection of damaged buildings in response to future events. The dataset and codes together with detailed explanations and visualization are made publicly available at \url{https://github.com/ya0-sun/PostEQ-SARopt-BuildingDamage}.

Figures (5)

• Figure 1: The SAR image coverage and zoomed-in views of three areas in the colored boxes in the SAR image, respectively.
• Figure 2: Examples of four study areas in the SAR image. Building footprint polygons before and after registration are plotted in red and green, respectively.
• Figure 3: The geocoding error from inaccurate height. (a) $H_t$ and $H_f$ are the accurate height and inaccurate height of a point, and $\theta$ is the incidence angle. The height error $\delta H$ results in an error of $\delta L = \delta H cos \theta$ in the slant range and $\delta G = \delta H cot\theta$ on the ground. (b) In the geocoded image, $\delta G$ is decomposed to $\delta x$ and $\delta y$ in the image coordinate system.
• Figure 4: Building signature in SAR imagery and building's geometric correspondence between SAR and building polygon: the near-range side of the building footprint corresponds to the double bounce line in the SAR image, which is the far-range side of the facade signaturessun2020auto.
• Figure 5: Examples of the dataset: a, b, c are intact buildings, and d, e, f are damaged buildings.