Illustrative: Belgian and U.S. Navy personnel conduct a sonar mine countermeasure drill during RIMPAC 2024. U.S. Navy photo by Petty Officer 3rd Class Iain Page via DVIDS (public domain). Used for illustration; it does not depict the ValidITy software. Appearance of U.S. Department of Defense imagery does not imply endorsement.

GEOMAR Open-Sources ValidITy: A GPU-Accelerated Tool for Detecting Munitions in Multibeam Sonar Data

GEOMAR's Deep Sea Monitoring Group has released ValidITy, an open-source tool that helps analysts annotate and detect seabed features, including unexploded ordnance, in multibeam echosounder bathymetry. Published in late July 2026 and licensed under the EUPL v1.2, it pairs a GPU-accelerated geomorphometric derivative engine with rule-based classification and trainable image-recognition models to speed underwater munition surveys.

Technical Summary

ValidITy is a raster annotation tool built to help analysts find and label features on the seabed. It works on digital elevation models, the gridded depth surfaces produced by multibeam echosounder (MBES) surveys. Rather than reading raw bathymetry, an operator computes geomorphometric derivatives from it: slope, aspect, positive and negative openness, the topographic position index (TPI), and the vector ruggedness measure (VRM). Each derivative sharpens a different kind of anomaly. A compact object standing proud of a flat seabed, such as an artillery shell, an aircraft bomb or a sea mine, tends to light up under the right combination of layers. The GEOMAR Deep Sea Monitoring Group developed the tool inside its ValidITy project, short for Validation of Intelligent Terrain feature recognition methods for hydrographic data, which ran from July 2023 to June 2025 under the Helmholtz Initiative and Networking Fund.

The release has two parts. The software itself, version 0.3.0, is now open source under the European Union Public Licence (EUPL) v1.2 and published on GitHub, with a formal software citation issued through GEOMAR's OceanRep repository. Alongside it, the group published a 16-page research report, Deep Sea Monitoring Reports DSMR-0004, documenting the GPU-accelerated engine that does the heavy arithmetic. Torsten Frey, the GEOMAR researcher who leads the group's work on munitions in the sea, said the project began when offshore survey contractors asked whether GEOMAR's munition-survey research could improve feature detection in their MBES data. Fugro Germany supported the work.

Benchmarked against the open-source SAGA GIS package, the engine computes the vector ruggedness measure and negative openness more than one hundred times faster, and the topographic position index by several orders of magnitude, turning a slow iterative task into real-time parameter tuning. GEOMAR Deep Sea Monitoring Reports DSMR-0004, July 2026

Analysis of Effects

The bottleneck ValidITy attacks is analyst time, not sonar hardware. Isolating a target in derivative space is iterative work. An operator adjusts a radius or a threshold, recomputes, looks, and adjusts again. On a high-resolution grid that loop can run to minutes per pass, which discourages the fine tuning that separates a munition from a boulder. The GEOMAR engine, written in Rust, uses a tiling scheme and GPU compute kernels through OpenCL so that it can process grids larger than the available graphics memory while redrawing derivatives as the operator moves a slider. The reported gains, more than a hundredfold for two of the most useful derivatives, are the difference between tuning by trial and tuning by eye.

On top of the derivative engine sit two ways to scale one interpreter's judgement across a large dataset. The first is a rule-based classification dictionary, where an operator encodes the derivative thresholds that define a feature of interest and applies them across a whole survey. The second is a local image-recognition model, a compact neural network the operator trains on their own annotations to extend a partial set of labels over an entire grid. GEOMAR is candid about the commercial history. The group has said that exploiting the tool as a product proved economically unfeasible, so it has opened the code to the community rather than let the work lapse.

Personnel and Safety Considerations

For explosive ordnance disposal (EOD) and survey practitioners, the important boundary is what bathymetry can and cannot tell you. An MBES survey measures the shape of the seabed. It does not measure what an object is made of, whether it holds an explosive fill, what that fill is, or the state of any fuze. A bright anomaly in a ruggedness layer is a candidate, not an identification. Confirming that a contact is ordnance, and then classifying it by hazard division (HD) and compatibility group (CG), still depends on the established survey chain: magnetometer and sub-bottom profiler data, then close visual inspection by remotely operated vehicle or diver, and finally a render-safe or disposal decision taken by qualified personnel. ValidITy sharpens the first step in that chain. It is not a target-identification system and it does not touch the disposal decision. Sea-dumped munitions are treated as inherently hazardous regardless of age, because corroded casings, degraded or sensitised energetic fills, and leaking chemical agent can all raise the risk of an item that looks quiescent on a sonar record.

Data Gaps

The released materials describe the tool and its engine but not its detection performance against verified ordnance. Several parameters that a munitions survey needs are not stated. The published examples emphasise boulders and natural seabed features rather than a validated munition signature library, so the probability of detection and the false-alarm rate against real unexploded ordnance are unknown. The minimum detectable object size relative to grid resolution is not given, which matters because MBES resolution sets a floor below which small-calibre items and partially buried ordnance may present too little relief to register. There is no independent benchmarking against a ground-truthed dumpsite in this release. And composition, fill and fuze state sit outside the reach of any bathymetric method, a limit of the physics rather than of the software.

Corrections & updates welcome. If you hold open-source data that refines or corrects any parameter in this article, please contact [email protected] citing the specific claim and your source. Verified corrections will be incorporated and credited in the revision history. AI-assisted technical assessment based on open-source material. Not a formal intelligence product.