Researchers at the University of Miami Rosenstiel School of Marine, Atmospheric, and Earth Science have demonstrated an airborne imaging system that can detect hidden underwater explosives with 100 percent accuracy, using a combination of drone technology and two NASA-developed optical systems.
The system, described in a study published in Frontiers in Marine Science, combines Fluid Lensing, a custom algorithm that removes wave distortion in real time to capture crystal-clear seafloor images, and MiDAR (Multispectral Imaging, Detection, and Active Reflectance), which shoots multiple wavelengths of light through the water column to illuminate dark ocean depths. Mounted on drones flown over the Florida Keys, the setup produced distortion-free multispectral imagery that an AI model then analyzed for signs of unexploded ordnance.
The results were unambiguous: the system identified every inert test munition and decoy placed on the seafloor around Broad Key, even after weeks of biological growth and sediment buildup had heavily obscured the targets.
A global problem hiding in shallow water
Decades of military conflict have left unexploded bombs, mines, and artillery shells scattered across shallow coastal waters worldwide. Many lie in waters shallower than 10 meters where conventional sonar is ineffective, the sonar platforms cannot operate in such shallow depths, and where wave distortion and shifting sand render standard optical cameras useless.
“Unexploded ordnance in shallow waters remains a serious global challenge,” said Ved Chirayath, Vetlesen Endowed Chair of Earth Sciences at the Rosenstiel School and lead author of the study. “Our results demonstrate a scalable, airborne solution that can help improve detection accuracy and support safer coastal environments.”
Existing methods rely on diver searches and acoustic boat surveys, both of which are slow, expensive, and dangerous. The old munitions leak toxic chemicals into fragile ecosystems and threaten tourists, shipping lanes, and coastal infrastructure projects across old European battlefields and Pacific dumping grounds.
How the AI finds bombs
The machine learning model was trained on high-resolution drone imagery to recognize the geometric signatures of munitions, distinguishing them from coral formations, rocks, and natural debris. Because Fluid Lensing removes wave distortion before the image reaches the AI, the model works with clean data rather than trying to compensate for blurry inputs, a design choice the researchers credit with the 100 percent detection rate.
The next step is testing across a wider range of marine environments, including murky Atlantic channels and deep Pacific bays, to validate whether the system maintains its accuracy under more challenging water conditions.