IRShips

IRShips is a dataset of synthetically generated IR imagery designed for the training and development of deep-learning based ATR algorithms.

Images were generated with CounterSim using a virtual camera in the 8-14 µm waveband.

The ships included in this dataset are:

Ships are imaged at:

• ship-camera distances of 1000, 1111, 1250, 1333, 1429, 1667, 2000, 2500, 3333, 5000, 10000 metres
• bearings from 0 - 359 degrees
• camera pitch of 0, 5 and 10 degrees

Setup

Run the setup.py script to unpack the dataset and to download and pre-process some images for use in online data augmentation.

Simply use the command:

python3 setup.py

NB this script will require:

• numpy
• PIL
• requests

The third-party images downloaded in this process are itemised in urls.yaml . For each image, the following information is given:

• image_url: a direct web address to the image
• reference: a reference for the image
• licence: the terms under which the image is licensed
• transforms: the image transforms applied during setup

Contents and Layout

At its top level, the IRShips dataset contains three directories and two files:

• images contains the 972,000 synthetic IR images (1024 x 512).
• labels contains the 108,000 binary masks (each corresponds to 9 IR images).
• augment contains the real-world data used for image augmentation.
• key.yaml contains a key that indexes the possible ship type and class designations.
• metadata.csv contains metadata for each synthetic image.

The augment directory contains the following three directories:

• clutter contains images relating to the augmentation of background clutter augmentation.
• sea contains contains images for sea-state augmentation from both an elevated and horizontal perspective.
• sky contains pre-processed images to be used for sky-state augmentation.

The sea directory contains the following two directories:

• elevated: images in here will be used for sea-state augmentation when the pitch of an image is greater than zero.
• horizontal: images in here be used for sea-state augmentation when the pitch of an image is zero.

The clutter directory contains directories, each of which contains images for background clutter augmentation. The name of these sub-directories denote the type of the clutter. The possible range of pixel intensities for each 'type' of clutter can be altered independently, using the clutter_intensity_range keyword argument of the Dataloader object.

The default clutter types are ice, landscape and structure, however it is possible to add any number of additional types .

Here is an overview of the structure described above:

irships
    ├ augment                 
    │    ├ clutter
    │    │    ├ ice
    │    │    │     ├ 0000.png
    │    │    │     ├ 0001.png
    │    │    │     └ ...
    │    │    ├ landscape
    │    │    │     ├ 0000.png
    │    │    │     ├ 0001.png
    │    │    │     └ ...
    │    │    └ structure
    │    │          ├ 0000.png
    │    │          ├ 0001.png
    │    │          └ ...
    │    ├ sea
    │    │    ├ elevated
    │    │    │     ├ 0000.png
    │    │    │     ├ 0001.png
    │    │    │     └ ...
    │    │    └ horizontal
    │    │          ├ 0000.png
    │    │          ├ 0001.png
    │    │          └ ...
    │    └ sky
    │         ├ 0000.png
    │         ├ 0001.png
    │         └ ...
    ├ images    
    │     ├ 00000000.png
    │     ├ 00000001.png
    │     └ ...             
    └ labels 
    │     ├ 00000000.png
    │     ├ 00000001.png
    │     └ ... 
    ├ key.yaml     
    └ metadata.csv   

Metadata

The metadata file, contains the following column names:

• filename - the name of the image file
• ship_type - the type of ship depicted in the image, such as Corvette, Frigate, etc.
• ship_class - the class of ship depicted in the image, such as Ada, Independence, etc.
• bearing - the bearing of the ship depicted in the image in degrees.
• range - the distance between the camera and the depicted ship in metres.
• pitch - the angle of pitch of the camera in degrees.
• signature - the integer identifier number of the ship thermal appearance (0-9).
• surface_temperature - the sea surface temperature in the image in degrees celsius.
• ambient_temperature - ambient temperature degrees celsius.
• x1
• y1
• x2
• y2 - the extents of the ship depicted in the image (normalised by image height and width).
• labelname - the name of the image's corresponding binary mask file.

Supporting code

Utils

Supporting code for this dataset can be found in utils.py and includes dataloader.

If you would like:

• more information
• to submit bug reports
• to submit a feature request
• the most up-to-date version

please visit my GitHub repository .

Example

The example.py script demonstrates how the dataloader may be used to load IRShips images.

Simply use the command:

python3 example.py

NB this script will require:

• cv2 (OpenCV)

Licence

Images included in this repository are licensed under CC BY-NC-SA 4.0, a full copy of this licence be found here .

Credit

If this dataset helps your research, please give recognition by citing the accompanying paper. Example BibTeX entry below:

@inproceedings{westlake2020deep,
  title={Deep learning for automatic target recognition with real and synthetic infrared maritime imagery},
  author={Westlake, Samuel T and Volonakis, Timothy N and Jackman, James and James, David B and Sherriff, Andy},
  booktitle={Artificial Intelligence and Machine Learning in Defense Applications II},
  volume={11543},
  pages={1154309},
  year={2020},
  organization={International Society for Optics and Photonics}
}