Radmantis reports 99% fish mix accuracy in hydropower test

Radmantis says a U.S. DOE-backed validation study showed its underwater AI platform could reconstruct trout species mix with 99% accuracy and detect fish with 95.5% reliability in dense lab conditions. The result points to faster, more precise fish passage monitoring for hydropower operators balancing clean power goals with ecosystem protections.

Why it matters: - Hydropower operators need better ways to monitor fish passage without slowing clean energy production. - Real-time species-level tracking can help protect migratory and endangered fish, manage invasive species and support environmental compliance. - The study suggests underwater AI could make fish monitoring more continuous and less dependent on manual counting.

What happened: - Radmantis completed a hydropower research validation study with the Hydropower Testing Network, ENERGYWERX, the U.S. Department of Energy’s Water Power Technologies Office and Verdantas. - The study was designed to test whether Radmantis could identify fish species composition in complex underwater scenes. - The company said the platform reconstructed fish species mix with 99% accuracy and detected fish with 95.5% reliability, measured as mAP50. - Researchers ran the validation at Verdantas Flow Labs’ Alden Campus hydraulics laboratory in Holden, Massachusetts.

The details: - The two-month study used varying combinations of rainbow trout and brown trout because the species look similar. - The setup tested the platform in dense group movement and visually complex conditions. - On a held-out validation set, the system reached 89.8% precision and 90.0% recall. - Across 40 single-blind trials, Verdantas knew the true species mix while Radmantis analyzed underwater video without that information. - The platform identified the exact brown-to-rainbow composition in 36 of 40 trials. - In the remaining four trials, the system missed the true mix by no more than five fish. - Dan Giza, Department Lead for Aquatic Resource and Bioengineering at Verdantas Flow Labs, said the collaboration adds data-driven tools for fish passage decisions and conservation efforts. - Robert Huber, Technology Lead at Radmantis, said smart underwater imaging can help make hydropower structures more sustainable and responsive to ecosystem needs.

Between the lines: - The test focused on trout because similar-looking species are a harder and more useful benchmark than simple counting. - High-accuracy species composition data matters because different hydropower sites face different ecological and regulatory obligations. - The same monitoring approach could help operators track native and migratory populations, flag invasive species and document compliance. - Radmantis is also applying related technology on the Colorado River to support invasive species control and native fish recovery.

What’s next: - Radmantis is seeking partnerships with hydropower operators and research institutions. - The company plans additional development and field validation across more species and operating environments. - The broader goal is to move the technology from controlled lab testing into real-world hydropower sites.

The bottom line: - The validation study gives Radmantis a strong proof point for using underwater AI to monitor fish passage in real time while supporting hydropower operations and river health.