Deutsch
 
Datenschutzhinweis Impressum
  DetailsucheBrowse

Datensatz

DATENSATZ AKTIONENEXPORT

Freigegeben

Zeitschriftenartikel

DC4Flood: A Deep Clustering Framework for Rapid Flood Detection Using Sentinel-1 SAR Imagery

Urheber*innen
/persons/resource/kasra

Rafiezadeh Shahi,  Kasra
4.4 Hydrology, 4.0 Geosystems, Departments, GFZ Publication Database, Deutsches GeoForschungsZentrum;

Camero,  Andrés
External Organizations;

Eudaric,  Jeremy
External Organizations;

/persons/resource/kreib

Kreibich,  H.
4.4 Hydrology, 4.0 Geosystems, Departments, GFZ Publication Database, Deutsches GeoForschungsZentrum;

Externe Ressourcen
Es sind keine externen Ressourcen hinterlegt
Volltexte (frei zugänglich)

5025871.pdf
(Postprint), 7MB

Ergänzendes Material (frei zugänglich)
Es sind keine frei zugänglichen Ergänzenden Materialien verfügbar
Zitation

Rafiezadeh Shahi, K., Camero, A., Eudaric, J., Kreibich, H. (2024): DC4Flood: A Deep Clustering Framework for Rapid Flood Detection Using Sentinel-1 SAR Imagery. - IEEE Geoscience and Remote Sensing Letters, 21, 4008605.
https://doi.org/10.1109/LGRS.2024.3390745


Zitierlink: https://gfzpublic.gfz-potsdam.de/pubman/item/item_5025871
Zusammenfassung
Severe flood losses have been on the rise, and this trend is expected to become increasingly prevalent in the future due to climate and socio-economic changes. Swiftly identifying flooded areas is crucial for mitigating socio-economic losses and facilitating effective recovery. Synthetic aperture radar (SAR) sensors are operational in all-weather, day-and-night conditions and offer a rapid, accurate, and cost-effective means of obtaining information for quick flood mapping. However, the complex nature of SAR images, such as speckle noise, coupled with the often absence of training/labeled samples, presents significant challenges in their processing procedures. To alleviate such hindrances, we can benefit from unsupervised classification approaches (also known as clustering). Clustering methods offer valuable insights into newly acquired datasets without the need for training or labeled samples. However, traditional clustering approaches are predominantly linear-based and overlook the spatial information of neighboring pixels during analysis. Thus, to attenuate these challenges, we propose a deep-learning (DL)-based clustering approach for flood detection (DC4Flood) using SAR images. The primary advantage of DC4Flood over existing DL-based clustering approaches lies in its ability to capture multiscale spatial information. This is achieved using multiple dilated convolutions with varying dilation rates and subsequently fusing the extracted multiscale information to effectively and efficiently analyze SAR images in an unsupervised manner. Extensive experiments conducted on SAR images from six different flood events demonstrate the effectiveness of the proposed DC4Flood. The code of the work will be available at https://github.com/Kasra2020/DC4Flood .