Datasets
Standard Dataset
Sentinel-2 Wildfire Change Detection (S2-WCD)
- Citation Author(s):
-
DimitrisValsamis
Centre for Research & Technology Hellas (CERTH)AlexandrosOikonomidisCentre for Research & Technology Hellas (CERTH)ChrysoulaChatzichristakiCentre for Research & Technology Hellas (CERTH)AnastasiaMoumtzidouCentre for Research & Technology Hellas (CERTH)IliasGialampoukidisCentre for Research & Technology Hellas (CERTH)StefanosVrochidisCentre for Research & Technology Hellas (CERTH)IoannisKompatsiarisCentre for Research & Technology Hellas (CERTH) - Submitted by:
- Dimitrios Valsamis
- Last updated:
- Fri, 12/06/2024 - 05:14
- DOI:
- 10.21227/2t8j-t191
- Data Format:
- Research Article Link:
- License:
42 Views- Categories:
- Keywords:
Abstract
Abstract
The Sentinel-2 Wildfire Change Detection (S2-WCD) dataset is a specialized collection of multispectral satellite images created to support the development and evaluation of wildfire change detection models. Comprising 41 image pairs with a resolution of 1066 × 1066 pixels, the dataset includes both pre-event and post-event scenes captured across Europe and Oceania. Each image pair is accompanied by pixel-level ground truth masks that identify areas affected by wildfires. The dataset covers 26 significant wildfire events from January 2021 to September 2023 in countries such as Australia, France, Greece, Italy, Portugal, and Spain, ensuring diverse geographical representation. By providing high-quality, accurately labeled data, S2-WCD facilitates the training of deep learning models, improving the accuracy and reliability of wildfire damage assessment and aiding effective disaster management efforts.
The repository provides the Sentinel-2 Wildfire Change Detection (S2-WCD) dataset. Each sample consists of .tif files representing spectral bands of a region before and after a wildfire event, as well as corresponding label files in .tif format. All relevant spectral bands are included, allowing users to combine them according to their specific requirements. The dataset is readily available for download without requiring any additional preprocessing steps. Comprising a total of 41 image pairs, users can freely divide the dataset into training and testing subsets based on their specific experimental needs, ensuring flexibility in model development. The diverse range of scenarios included in the dataset helps prevent bias toward any single case, promoting the creation of robust and generalizable wildfire change detection models.
