Raw/L0 Multispectral Ship Classification Datasets

VDS2Raw - Classification

The VDS2Raw dataset comprises 64x64 pixel crops of raw granules derived from Sentinel-2 products, focusing specifically on vessels in Danish coastal areas. Each crop is centered on a ship and extracted from the original Sentinel-2 images to support vessel classification tasks. This dataset leverages Sentinel-2’s multispectral data, covering bands B2, B3, B4 and B8 with a spatial resolution of 10 meters, enabling coarse-grained vessel classification across three categories: Cargo, Sailing & Pleasure (S&P), and Fishing.

The dataset is structured with a total of 106 Cargo, 99 S&P, and 123 Fishing vessels, split into training (64 per class) and test subsets (42 Cargo, 35 S&P, and 59 Fishing vessels). 

VDVRaw - Classification

The VDVRaw dataset includes 64x64 pixel crops from VENμS satellite imagery, each centered on a maritime vessel, with high-resolution (5.3 meters at Nadir) imagery across 12 spectral bands. Each crop is taken from coregistered images, aligned to band B5 to ensure consistent spatial alignment, providing a robust foundation for classification tasks with detailed spectral data. The dataset targets two classes derived from AIS data: Bulk Carrier and Container Ship.

For the VENµS dataset, Bulk Carrier and Container Ship classes are represented with 643 and 398 samples, respectively. The dataset is structured into training (318 per class), and test (325 Bulk Carrier and 80 Container Ship vessels).