Self-supervised Learning for Building Damage Assessment from Large-Scale xBD Satellite Imagery Benchmark Datasets

Xia, Zaishuo; Li, Zelin; Bai, Yanbing; Yu, Jingquan; Adriano, Bruno

doi:10.1007/978-3-031-12423-5_29

Cited by 8 publications

(3 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most previous studies in this area fall into the categories of image segmentation [11,12] or object detection [13]. The insights offered include but are not limited to, the effectiveness of "attention mechanisms" and visual transformers built on those mechanisms [14,15,16,17,18], novel convolutional blocks [19] and even graph neural networks that account for dependencies between structure types and their damage conditions [20]. Other studies use additional data-for example, public building footprint inventories such as OpenStreetMap 2 for building localization [21], or as additional input data channels [22,23]-, or use both pre-and post-event images for damage identification through change detection strategies [24].…”

Section: Introductionmentioning

confidence: 99%

Deep Ensemble Learning for Rapid Large-Scale Post-Earthquake Damage Assessment---Application to 2023 Kahramanmaraş Earthquake Sequence

Soleimani-Babakamali,

Askari,

Heravi

et al. 2023

Preprint

View full text Add to dashboard Cite

Extensive impacted region and government damage surveys following the 2023 Kahramanmaraş earthquake sequence provided a unique opportunity to design and validate a rapid post-event damage assessment framework with satellite images, addressing the typical challenges (e.g., generalization to unseen areas) encountered in rapid regional assessment efforts. By analyzing images from the significantly impacted city of Antakya, we manually identified visual features and compared those with damage states in the survey. Then, an Artificial Intelligence-based framework was developed to automate segmentation and damage assessment processes for future test areas. The resulting tool, dubbed RAPID-A, is an uncertainty-aware ensemble of various deep segmentation models fed with public data released shortly after such incidents. Furthermore, the extent of the damages reported in the survey allowed us to propose and validate a strategy to identify various intensities of damage using a single model, considering the less severe damage as more uncertain. Test case studies in Gaziantep and Kahramanmaraş demonstrate that RAPID-A is generalizable, accurate, and efficient. The significance of this framework lies in its ability to assist in rapid initial damage evaluations before expert dispatch or to replace it in countries that lack the capacity to launch large-scale regional assessment programs to guide foreign support.

show abstract

Section: Introductionmentioning

confidence: 99%

Deep Ensemble Learning for Rapid Large-Scale Post-Earthquake Damage Assessment---Application to 2023 Kahramanmaraş Earthquake Sequence

Soleimani-Babakamali,

Askari,

Heravi

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Furthermore, the building type data generally have low spatial resolution (mostly administrative level one or zero 14 ) and is often biased toward certain regions of the world 11 . Recent research has also applied deep-learning models to predict building damage by comparing satellite image data before and after floods, earthquakes and conflict, using purely data-driven deep learning models [15][16][17][18] . This requires satellite image data, where the models in these (pre-print only) articles ( [15][16][17][18] ) were trained on commercial (non-publicly available) satellite data.…”

Section: Introductionmentioning

confidence: 99%

“…Recent research has also applied deep-learning models to predict building damage by comparing satellite image data before and after floods, earthquakes and conflict, using purely data-driven deep learning models [15][16][17][18] . This requires satellite image data, where the models in these (pre-print only) articles ( [15][16][17][18] ) were trained on commercial (non-publicly available) satellite data. This approach limits the outcome predictor to building damage classification, and not mortality or displacement.…”

Section: Introductionmentioning

confidence: 99%

Data-Driven Earthquake Multi-Impact Modelling: A Comparison of Models

Patten,

Loake,

Steinsaltz

2023

Preprint

View full text Add to dashboard Cite

A broad range of supervised Machine Learning and parametric statistical, geospatial and non-geospatial models are applied to model both aggregated observed impact estimate count data and satellite image-derived geo-located building damage labelled data, via regression- and classification-based models, respectively. For the former, models are ranked via predictive performance of mortality, population displacement, building damage and building destruction for 161 earthquakes in 61 countries. For the latter, models were ranked via classification performance (damaged/unaffected) of 369,813 geo-located buildings for 26 earthquakes in 15 countries. k-fold, 3-repeat cross validation is utilised to ensure out-of-sample predictive performance. Feature importance on several variables used as proxies for disaster-risk vulnerability indicates covariate utility. The 2023 Türkiye-Syria earthquakes event is shown to be unlike any of the other 161 earthquakes, resulting in high prediction error for the aggregated impact models compared to all other events, reflecting model limitations on extreme events. However, training the building damage classifier AdaBoost model on all except the 2023 Türkiye-Syria earthquakes, predictions on the 27,032 held-out buildings have an AUC of 0.93. Therefore, without any geospatial, building-specific or direct satellite image information, this model accurately classifies building damage, with significantly improved performance over satellite-image trained models found in the literature.

show abstract

Improving Disaster Response with U-Net and Attention Mechanism in Earthquake Damage Assessment Using Remote Sensing Data

Anvitha,

Maheswari

2024

Lecture Notes in Electrical Engineering

View full text Add to dashboard Cite

Self-supervised Learning for Building Damage Assessment from Large-Scale xBD Satellite Imagery Benchmark Datasets

Cited by 8 publications

References 15 publications

Deep Ensemble Learning for Rapid Large-Scale Post-Earthquake Damage Assessment---Application to 2023 Kahramanmaraş Earthquake Sequence

Deep Ensemble Learning for Rapid Large-Scale Post-Earthquake Damage Assessment---Application to 2023 Kahramanmaraş Earthquake Sequence

Data-Driven Earthquake Multi-Impact Modelling: A Comparison of Models

Improving Disaster Response with U-Net and Attention Mechanism in Earthquake Damage Assessment Using Remote Sensing Data

Contact Info

Product

Resources

About