Rotated Ship Detection Dataset in SAR Images    2022 No.1

Guest Editor: Xu Congan, Su Hang, Yan Wenjun

Rotated Ship Detection Dataset in SAR Images (RSDD-SAR) was built based on the domestic GF-3 satellite and the ESA TerraSAR-X satellite. This dataset consists of 84 scenes GF-3 data slices, 41 scenes TerraSAR-X data slices, and 2 scenes uncropped large images, includes 7,000 slices and 10,263 ship instances of multi-observing modes, multi-polarization modes and multi-resolutions, and has the characteristics of arbitrary rotation direction, large aspect ratio, high proportion of small targets and rich in scenarios. Figure.1 shows an example of the typical scenarios in RSDD-SAR.

dfe5a33c7bd64ec68ef443aa25c60cd8.png

Fig.1 Typical scenarios in RSDD-SAR

The size of slice in RSDD-SAR is unified to 512×512 pixels, the format of slice is a 3-channel, 24-bit, grayscale .JPG images, and the annotation of slice is a .XML format file. The annotation file records the polarization mode, resolution, rotated bounding box annotation and other information. The rotated bounding box annotation adopts the long-edge definition method, including the center point coordinate, long edge, short edge, and angle. For an example of the annotation file, see the ‘RSDD-SAR Dataset Instructions’. 2 scenes uncropped images and their annotations are the same as the slice.

For details of the dataset, you can see  “RSDD-SAR Dataset Instructions”.

Citation format: XU Congan, SU Hang, LI Jianwei, et al. RSDD-SAR: Rotated ship detection dataset in SAR images[J]. Journal of Radars, in press. doi: 10.12000/JR22007.


Data Usage Protocol of Journal of Radars

The data can be used free of charge for scientific research, teaching and so on, but the data source should be marked in the reference according to the citation format.

Data use for commercial purposes requires permission from the Editorial Department of Journal of Radars.