MASATI dataset (v2) - MAritime SATellite Imagery dataset


This dataset provides maritime scenes of optical aerial images from visible spectrum. The MASATI dataset contains color images in dynamic marine environments, and it can be used to evaluate ship detection methods. Each image may contain one or multiple targets in different weather and illumination conditions. The datasets is composed of 7,389 satellite images labeled according to the following seven classes: land, coast, sea, ship, multi, coast-ship, and detail. In addition, labeling with the bounding box for the location of the vessels is also included.

The next table shows the sample distribution for each class:

Main class Sub-class #samples Description
Ship Ship 1027 Sea with a ship (no coast).
Detail 1789 Ship details.
Multi 304 Multiple ships.
Coast & ship 1037 Coast with ships.
Non-ship Sea 1022 Sea (no ships).
Coast 1132 Coast (no ships).
Land 1078 Land (no sea)

For evaluation we have defined three additional sets by grouping samples of several classes as follows:

  • Set 1: Ship on high sea and ocean or high sea without ship.
  • Set 2: Set 1 plus two new subsets: ship on sea close to coast (then coast is visible), and coast (sea scene with coast visible but without ship).
  • Set 3: Set 2 plus three new subsets: ship image acquired at lower altitude compared with the set 1, land (inland this is without coastal areas), and multi (multiple instances of ships).

The dataset has been compiled between March and September of 2016 from different regions in Europe, Africa, Asia, the Mediterranean sea and the Atlantic and Pacific oceans.

LICENSE AND ACCESS


This dataset is shared only for non-profit research or educational purposes. If you use this dataset or a part of it, please respect these terms of use and reference the original work in which it was published.

All data were obtained from Microsoft® Bing™ Maps. You can consult the Bing Maps terms of use at https://www.microsoft.com/maps/product/terms.html. Please read carefully the included file with the terms of use shown in Microsoft® Bing™ Maps.

FILE FORMAT


The satellite images were acquired from Bing Maps in RGB and with different sizes, as size is dependent on the region of interest to be registered in the image. In general, the average image size has a spatial resolution around 512 x 512 pixels. The images are stored as PNG where pixel values represent RGB colors. The distance between targets and the acquisition satellite has also been changed in order to obtain captures at different altitudes.

To label the category of each image, an organization divided into folders was used, where each folder represents a category.

In addition, the bounding box with the exact location of each ship in the image is included. This labeling is stored in XML files using the annotation format of PASCAL VOC. For this, we used the tool "LabelImg" (https://github.com/tzutalin/labelImg).

RELATED PUBLICATIONS (CITATION)


Please, if you use this dataset or part of it, cite one or both of the following publications:

  @article{Gallego2018,
    author  = {Antonio-Javier Gallego, Antonio Pertusa and Pablo Gil},
    title   = {Automatic Ship Classification from Optical Aerial Images
                 with Convolutional Neural Networks},
    journal = {Remote Sensing},
    volume  = {10},
    number  = {4},
    year    = {2018},
    ISSN    = {2072-4292},
    doi     = {10.3390/rs10040511}
  }

  @article{Alashhab2019,
    author  = {Samer Alashhab, Antonio-Javier Gallego, Antonio Pertusa and Pablo Gil},
    journal = {IEEE Access},
    title   = {Precise Ship Location With CNN Filter Selection From Optical Aerial Images},
    year    = {2019},
    volume  = {7},
    pages   = {96567-96582},
    doi     = {10.1109/ACCESS.2019.2929080}
  }

This work was funded by the Spanish Government-Ministry of Economy, Industry and Competitiveness trough the projects RTC-2014-1863-8 and INAER4-14Y(IDI-20141234).

DOWNLOAD


To download this dataset fill in the following form:


https://goo.gl/forms/xrBmFfbSuwpyb3wS2



VERSIONS


V1 - Released in February 2018. This version of the dataset contained 6,212 satellite images classified in the same seven classes. The location of the boats was not included.

V2 - Released in June 2019. 1,177 new images were added (making a total of 7,389 images). The bounding box was added with the location of the boats.