NASTaR: NovaSAR Automated Ship Target Recognition Dataset

The NovaSAR Automated Ship Target Recognition (NASTaR) comprises 3415 ship patches extracted from NovaSAR S-band imagery, with labels aligned to AIS (Automatic Identification System) data. Key features of the dataset include: 23 distinct ship classes, separation between inshore and offshore samples, and an auxiliary wake dataset containing 500 patches where ship wakes are visible. Ship patches are provided in size of 512×512 pixels. Wake patches are available in both 512×512 and 1024×1024 pixels to offer higher spatial coverage. Each sample includes: GeoTIFF float32 images with backscatter information, and GeoTIFF uint8 images for visualization purposes. The corresponding ground-truth metadata includes ship type, geographic location, speed over ground, distance from shoreline, heading angle, width, height, and other relevant attributes. The dataset is primarily designed for ship type classification, but the provided ground-truth information also makes it suitable for detection tasks. Further description of the dataset, together with links to relevant code for benchmarking and analysis is available at https://doi.org/10.48550/arXiv.2512.18503.