ApertureLab · Synthetic Aperture Sonar Simulator

The seafloor is dark.<br>It doesn't have to be.

ApertureLab is a synthetic aperture sonar simulation and beamforming<br>workbench; built to accelerate the AI systems that will map and<br>understand the ocean floor at scale.

See the software in detail &rarr;

Hover for labels · click for full hi-res &rarr;

Isaac D. Gerg, Ph.D.

AI Scientist, ClimateAI ·<br>gergltd.com

Most underwater AI projects fail at the same seam: the ML team does not<br>understand the acoustics, and the sonar team does not understand deep learning.<br>I work at both ends. Twenty years of research spanning the complete signal<br>chain (acoustic wave scattering and IQ time-series recording, through<br>beamforming and image formation, to deep embeddings for ATR, segmentation,<br>image compression, and making each stage robust to domain shift) means I<br>can follow the signal from the seafloor to a model embedding and identify<br>exactly where it breaks.

ApertureLab is what that understanding looks like as a piece of software.

Wave Scattering Physics<br>IQ Time-Series Processing<br>TDBP & ω-k Beamforming<br>Image Formation<br>Deep Embeddings<br>ATR<br>Segmentation<br>Image Compression<br>Physics-Based AI/ML<br>Domain Robustness

What ApertureLab unlocks.

SAR foundation models exist because satellite imagery<br>was cheap, abundant, and labeled. Sonar data requires ships, dive operations,<br>and classified access. ApertureLab generates physics-accurate labeled sonar<br>data at satellite-imagery scale; the missing precondition that makes the<br>following possible for the first time.

Synthetic Datasets<br>Million-image labeled datasets from a single workstation

Every simulated image carries ground-truth labels derived automatically<br>from the scene: object class, position, orientation, burial depth, seafloor<br>type, shadow mask. A dataset that would require decades of ship time to<br>collect in the field can be generated overnight.

Object detection and segmentation labels

Pixel-accurate shadow and highlight masks

Full domain variation: depth, range, bottom type, aspect, burial

Rare-event coverage impossible to collect in the field

The data problem for underwater AI is a logistics<br>problem. ApertureLab is the logistics solution.

Foundation Models<br>No general-purpose foundation model has been trained on SAS imagery. Yet.

Not because the architecture does not exist (ViT, MAE, and contrastive<br>pretraining are all mature), but because training data at the necessary<br>scale never existed. ApertureLab closes that gap with parametric scenes<br>spanning the full space of acoustic environments.

Sim-to-real transfer for autonomous underwater vehicle perception

Environment-agnostic feature representations across seabed types

Edge-case and rare-target coverage unavailable in field collections

The sonar analogue to ImageNet-scale pretraining

This is the next rung. The ladder already exists.

Vision-Language Models<br>VLMs already respond to sonar. Fine-tuning is the obvious next step.

The 2026 IGARSS work shows VLMs classify SAS targets at 0.946 AUC<br>using only a text prompt describing highlight-shadow geometry, with zero<br>domain-specific training. Scene captions generated at render time make<br>the fine-tuning step on a million labeled ApertureLab images straightforward.

Auto-generated natural-language captions paired with every image

Query sonar archives by English phrase

Caption imagery for non-expert operators and analysts

A path to zero-shot object recognition across unseen target types

The zero-shot baseline is 0.946 AUC. Domain<br>fine-tuning is the straightforward next step.

Evidence

From one English sentence to a labeled dataset.

A single English prompt to ApertureLab, written through Claude Code: &ldquo;create a 60 by 100 m scene with a grid of different seafloor backgrounds and a target in the middle of each one, so I can crop 256 by 256 chips of the same object on different bottoms for ML training.&rdquo; The image above is what came back, rendered overnight on one workstation; physics through the entire chain. Sixty seafloor targets, ten varied bottom classes — clean sand, mud, rock, gravel, coarse gravel, silt, rippled sand, two rocky-sediment variants, and shelly sand hash — with roughness and reflectivity jittered per cell so no two cells match. Every green box, name, and seafloor zone outline is generated automatically from the scene file at simulation time, not hand-annotated. The three insets along the right show the same target lifted off three of those backgrounds.

The same physics that renders a labeled scene forward is what trains the model that reads one backward.

Simulate

Scene editor for sonar geometry, seafloor types, objects, and<br>platform motion. GPU ray-trace physics backend for accurate<br>acoustic scattering and shadow geometry.

Beamform

Integrated TDBP and ω-k pipelines produce georeferenced SLC<br>imagery directly in the tool.

Inspect

Interactive viewer with physical-coordinate cursor readout,<br>three output variants, pan/zoom.

Why synthetic...