-
Notifications
You must be signed in to change notification settings - Fork 3
/
Dataset description
10 lines (10 loc) · 1.3 KB
/
Dataset description
1
2
3
4
5
6
7
8
9
10
VHRShips
The success in deep learning (DL) world is obvious nowadays and getting more popular with new released application areas. Working on optical satellite image is one of them.
Large areas, numerous and complex content are some of the major reasons make people do research in the remote sensing images with DL.
Detection, Localization, Recognition and Identification (DLRI) of ships from optical satellite images with DL is one of these research areas.
DLRI of ships has many usages like monitoring maritime traffic, using in the defense purposes, preventing illegal immigration, smuggling, over-fishing, sea pollution,
protecting sea mammals and others. Even developing an efficient and successful DL algorithm is very important for a practicable DL application,
working with an appropriate dataset is critical. This study proposes a new dataset – Very High Resolution Ships (VHRShips) – to train, validate and test DL algorithms in DLRI
of ships from optical satellite images. VHRShips comprises of 6312 optical satellite images collected from Google Earth. While 1000 images don’t include any ship, rest of them
consist 11337 ships in 34 different classes. VHRShips has high definition, inshore and offshore images from various locations, rich metafile content and includes noisy images
representing real scenarios.