Safe Pro completes its Air Force FOD dataset contract, but gives no details of the data
Safe Pro Group (Nasdaq: SPAI), whose AI software searches drone imagery for landmines, cluster munitions and unexploded ordnance, said on 29 October that it has completed all research milestones under a Phase I Small Business Innovation Research contract from AFWERX, the innovation arm of the Department of the Air Force. The contract, announced on 13 July, was to develop a new AI dataset from images of the foreign object debris commonly encountered on US Air Force and civilian airfields.
According to the company, the work consisted of three things: demonstrating its SpotlightAI computer vision software on small-object detection, collecting and analysing high-resolution FOD imagery, and preparing technical project reports. It intends to pursue further SBIR-funded development, which it says is “subject to review and determination by AFWERX”.
What neither announcement gives is any detail of the data. There is no image count, no list of object classes, no airfield named as the place the imagery was collected, no word on whether the debris was found in operations or placed for the camera, and no contract value.
Those are the details that matter, because data is the hard part. This site argued last year that the real constraint on AI FOD detection is that nobody has the data: debris on an operational runway is rare, varied and removed as soon as it is found, so detectors end up trained on staged objects or synthetic images. A funded effort to build a FOD image dataset is a direct attempt at that problem. It is not the only military interest in the field, either: the Navy’s Direct to Phase II topic seeking persistent airfield FOD and pavement inspection closed to proposals on 23 September.
It is also worth noting where Safe Pro is coming from. Its detection models are trained, it says, on one of the world’s largest real-world drone-based imagery datasets, and it presents the airfield contract as a way to take its computer vision beyond battlefield and post-conflict land reclamation. Whether software built to find mines in a field can also find a bolt on a grooved runway at night is exactly the question a dataset like this one should be able to answer.
For airports and anyone buying detection systems, nothing here is yet a product. In SBIR terms Phase I is the feasibility stage, and the stated deliverable is technical reports; the release does not say whether the dataset will be made available beyond the Air Force. The test of it is the one set out in that 2025 piece: trained on what, photographed where, and in what weather. Safe Pro has not yet said.
Sources (GlobeNewswire, 29 October, GlobeNewswire, 13 July)