Radmantis AI Accurately Identifies Fish Species in Hydropower Study

Radmantis, a blue tech company, has developed an AI underwater vision platform that accurately identifies fish species in dense groups, marking a significant advancement in hydropower monitoring. The platform, which uses embedded vision, AI-based inference, and industrial automation, achieved a 99% accuracy rate in a validation study conducted by Verdantas Flow Labs, a leader in hydraulic engineering and fish passage design.

The study, supported by the U.S. Department of Energy's Water Power Technologies Office (WPTO) through the Hydropower Testing Network (HyTN), involved the detection and classification of brown trout and rainbow trout. Researchers at Verdantas Flow Labs used the platform to track and identify fish moving in a dense group during a two-month validation study at the Alden Campus hydraulics laboratory in Holden, Massachusetts. The platform successfully distinguished between the two species, achieving 95.5% detection performance (mAP50), with 89.8% precision and 90.0% recall.

Across 40 single-blind trials, the platform reconstructed the brown-to-rainbow trout mix with 99% accuracy. Verdantas Flow Labs knew the actual species composition, while Radmantis independently analyzed the underwater video without that information. The platform identified the true composition exactly in 36 of 40 trials and to within 5 fish in the remaining trials. This high level of accuracy demonstrates the platform's capability to provide clear, continuous insight into fish passage, which is crucial for balancing clean energy production with ecological and regulatory requirements.

Radmantis' monitoring platform addresses the challenge of balancing energy production with ecological and regulatory requirements by providing real-time analytics. The platform can support environmental compliance, population assessment, native species preservation, and invasive species control. The technology has applications beyond fish counting, extending to the identification of species within dense, visually complex environments.

The ability to identify species within dense, visually complex environments extends beyond fish counting. Each hydropower site has different species of concern, and continuous monitoring can provide operators with more actionable data to manage native and migratory populations, identify invasive species, and meet environmental requirements. Radmantis is also applying related technology on the Colorado River in support of invasive species control and native fish recovery efforts.

With the HyTN validation study complete, Radmantis is seeking partnership opportunities with hydropower operators and research institutions to continue development and field validation across additional species and operating environments.