MASATI dataset - 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 6212 satellite images labeled according to the following seven classes: land, coast, sea, ship, multi, coast-ship, and detail. The next table shows the sample distribution for each class:

Main class Sub-class #samples Description
Ship Ship 1015 Sea with a ship (no coast).
Detail 1789 Ship details.
Multi 188 Multiple ships.
Coast & ship 121 Coast with ships.
Non-ship Sea 1010 Sea (no ships).
Coast 1054 Coast (no ships).
Land 1035 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).

We plan to continuously collect and upload new marine scenes. As researchers use the data, we will list results and benchmarks here. If you have any results on the data that you would like to be listed here, please contact us (jgallego AT ua DOT es)

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.


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 Please read carefully the included file with the terms of use shown in Microsoft® Bing™ Maps.


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.


Please, if you use this dataset or part of it, cite the following publication:

    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}

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).


To download this dataset fill in the following form: