Water_dataset_v0

General Information : 

The European Space Agency (ESA) initiated the OrbitalAI Φsat-2 challenge to establish a robust network of applications for Earth observation that capitalize on edge computing technologies. In participation in this challenge, a simulation tool known as the PhiSat-2 simulator was utilized. This simulator facilitated the recreation of the L1C reflectance product from the Sentinel-2 Multispectral sensor, featuring a Ground Sampling Distance (GSD) of 10 meters. Through the application of bicubic interpolation, the resultant simulated product achieved a heightened resolution of 4.75 meters GSD.

Subsequently, this newly generated high-resolution product was employed to generate training data tailored specifically for the detection of water-inundated roads. The focus was directed towards Bengaluru city on three distinct dates: May 3rd, September 5th, and October 25th, all occurring in 2022.

Training Data preparation : 

Simulated satellite imagery is utilized to compute the Normalized Difference Water Index (NDWI) for the identification of areas containing water bodies. By applying a threshold of 0.01, a binary mask is generated to delineate water from land. To enhance the accuracy of this mask and subsequently refine the overall machine learning model, a visual inspection and reclassification step is conducted using QGIS, an open-source software. Subsequently, the resulting water mask, sized at 4096 x 4096 pixels, is utilized to generate smaller image chips measuring 256 x 256 pixels each. A total of 768 image chips are extracted from the water mask to facilitate the development of the machine learning model.

Description of Water_dataset_v0: 

The dataset, titled "Water_dataset_v0," comprises information regarding water bodies in satellite imagery and is structured into two subfolders:

images: This directory houses the actual satellite image chips. Each image chip is formatted as a TIFF file and possesses dimensions of 256 x 256 pixels.

annotations: Within this folder are the corresponding water mask image chips associated with each image chip found in the "images" directory. These water masks also maintain a dimension of 256 x 256 pixels. of 256 x 256 pixels and are saved in TIFF format. The pixel values in the water mask represent the presence of water: 0 represents non-water pixels and 1 represents water pixels.

Example File Naming:

Images named using a consistent scheme like "790620-1430950_32643_PHISAT2-BANDS-GRID_0_2022-05-03T05-25-11_000_chip_0_0.tiff" "790620-1430950_32643_PHISAT2-BANDS-GRID_0_2022-05-03T05-25-11_000_chip_0_1.tiff"

Corresponding water masks would then be named "790620-1430950_32643_PHISAT2-BANDS-GRID_0_2022-05-03T05-25-11_000_chip_0_0.tiff""790620-1430950_32643_PHISAT2-BANDS-GRID_0_2022-05-03T05-25-11_000_chip_0_1.tiff"