ApertureLab synthetic aperture sonar datasets

Detection and segmentation models for a new kind of imagery are trained on large collections of labeled examples, and for synthetic aperture sonar such collections are scarce. The two datasets on this page each contain 1000 swaths from a simulated HISAS 1030-class sonar at 2.5 cm resolution, labeled in COCO format. Survey data are scarce because they are expensive to acquire, much of what exists cannot be distributed, and every object in every image must be labeled by a trained analyst. In simulation the position and extent of each object are known when the image is formed, so the labels are computed from the scene description instead of drawn by hand.

Both datasets were simulated with ApertureLab and are published on Hugging Face under the ALSD-1.0 license. Each swath is accompanied by the scene file that produced it, so that any swath can be regenerated or modified. The labels serve detection and segmentation work; the scene files, and in version 1.1 the terrain height maps, allow an image to be compared with the seafloor and objects that produced it. Version 1.1 corrects the label geometry of version 1.0, adds up to 5 m of terrain relief, and supplies a height map and a boulder table with each swath. Version 1.0 remains available as published.

ApertureLab Synthetic SAS Dataset v1.1

Version 1.1 has the same swath footprint and sonar as version 1.0, with every scene drawn from a new master seed. Each label polygon is reduced by the burial of its object and projected into the image through the terrain height beneath it. The seafloor carries up to ±5 m of relief, and each image is a georeferenced GeoTIFF with nothing rendered over it. The swaths tile edge to edge northward from the origin of a local transverse Mercator frame, so that the release forms a single strip 80 km long and 200 m wide. Each kind of object is its own label category.

The label categories are:

wedge_stl, a wedge resting level with its base sunk into the bed;

wedge_belly, the same wedge lying on its flat underside;

truncated_cone, a truncated cone standing on its base;

cylinder, a cylinder lying on its side;

rock, every boulder of 0.75 m or more, under the supercategory clutter;

twelve parametric clutter shapes, pyramid to torus, and ten clutter meshes, tire to rock_pile, each its own category under clutter.

Six sonar chips in a row, each with its label polygon drawn and its category name below: wedge_stl, wedge_belly, truncated_cone, cylinder, a placed clutter object and a boulder (rock) — **One example of each class**6 m chips at the native 2.5 cm, each cropped around its COCO box with the footprint polygon drawn; clutter appears twice, as a placed object and as a boulder

A whole v1.1 sonar swath of rippled sand, along-track up and range to the right — **The image**a rippled sand scene; the palette GeoTIFF shown through its own palette, 200 m of ground range by 80 m along track, reduced from 8000 by 3200 px

A 40 m by 20 m window of a rocky swath at full resolution with target, clutter and boulder polygons drawn — **A window, labeled**40 m by 20 m of a rocky scene at full resolution with its COCO file drawn: targets in green, clutter objects in magenta, boulders in orange, shadow polygons dashed

Instrument

The instrument is a simulated HISAS 1030-class synthetic aperture sonar with a 100 kHz center frequency, 30 kHz of bandwidth and 32 receive channels. It images the starboard side from a straight track, at an altitude drawn per swath from 15 to 25 m.

Size

The dataset contains 1000 swaths, split 800 train, 100 validation and 100 test, covering 4.66 nmi² (16 km²) of seafloor. It holds 2000 targets, 500 of each target class, over 700 swaths, and 300 swaths contain no target. Label counts by category and split are tabulated on the v1.1 page.

Image

Each image is 8000 by 3200 px at 2.5 cm, covers 200 m of ground range by 80 m along track, and is stored as an 8-bit palette GeoTIFF with no watermark.

Per swath

Each swath comes with the drc.tif image, the coco.json labels and acquisition record, the dem.png height map, the rocks.csv boulder table and the scene YAML.

Labels

The labels fall in 27 COCO categories: four targets under the supercategory target, and rock plus 22 clutter shapes and meshes under clutter. Category id 28, clutter_other, is reserved for scenes outside the dataset and never occurs in it.

License

The dataset is released under ALSD-1.0, the ApertureLab Synthetic SAS Dataset License.

ApertureLab Synthetic SAS Dataset v1.0

Version 1.0 contains 1000 swaths over five seafloor families, with truncated cones and wedges as the two target classes and 22 classes of clutter. Each image is labeled with COCO polygons and accompanied by its scene file. The release is kept available as published. Its one known defect is an information box rendered into the lower-right corner of every image; the labels beneath it are correct.

A whole v1.0 sonar swath of rippled sand with trawl scars, along-track up and range to the right — **The image**scene si_0008, ripple fields on sand with trawl scars; reduced from 8000 by 3200 px; the information box is in the lower-right corner

A 50 m by 30 m window of a v1.0 swath at full resolution with green target polygons and a cyan clutter polygon — **A window, labeled**scene si_0581, 50 m by 30 m at full resolution with its COCO polygons: targets in green, clutter in cyan

The same 50 m by 30 m window without labels — **The same window, clean**the window above without its labels

Twelve 10 m chips of cones and wedges on sand, gravel, silt, rock and mud, each with its green label polygon — **Target chips**a cone and a wedge on each substrate, near and far range, exposed and buried; 10 m chips at the native 2.5 cm

Instrument

The instrument is a simulated HISAS 1030-class synthetic aperture sonar with a 100 kHz center frequency, 30 kHz of bandwidth and 32 receive channels, on a straight track at 20 m altitude.

Size

The dataset contains 1000 swaths, split 800 train, 100 validation and 100 test, covering 4.66 nmi² (16 km²) of seafloor in about 40 GB of images. It labels 2000 targets and 2479 clutter objects.

Image

Each image is 8000 by 3200 px at 2.5 cm, covers 200 m of ground range by 80 m along track, and is stored as RGB TIFF with LZW compression.

Per swath

Each swath comes with the drc.tif image, the coco.json labels and the scene YAML; manifest.csv, objects.csv and coco_all.json cover the whole set.

Labels

The dataset holds 5503 COCO polygons in 27 categories: the two target classes, 22 clutter classes and three seafloor features.

License

The dataset is released under ALSD-1.0, the ApertureLab Synthetic SAS Dataset License.