U.S. flag

An official website of the United States government

Dot gov

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Https

Secure .gov websites use HTTPS
A lock () or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

Breadcrumb

  1. Home
  2. Evaluation

Efficiency and accuracy assessment of an automated image classifier for commercially important fish species

Published Date
Fiscal Year
FY 2026
Bureaus and Offices
National Oceanic and Atmospheric Administration
Program
National Marine Fisheries Service

Experts that analyze video collected from underwater optical systems can routinely provide accurate species counts. However, the high opportunity costs of expert time have entailed delays in processing large amounts of video data. Artificial Intelligence (AI) has the potential to reduce these delays and costs. To establish a baseline for this potential, we conducted a formative evaluation of a model used to count fish by species, run in the Visual and Image Analytics for Multiple Environments (VIAME) software. We compared the results from an application of the software against those generated by human experts across 357 Gulf of America underwater video deployments. The tool excelled at minimizing false-positive identifications. However, it consistently undercounted target species that were actually present. The missed identification was prevalent, particularly in cases where there was a high-degree of species diversity within the video. Automated processing time was substantially longer than the time spent for manual review. Fish density and species diversity increased processing time for both the manual reviewer and the automated classifier. In general, the statistical explanation for variations in processing time for manual review was more robust than the VIAME tool. We conclude that operating multiple continuously running machines and improving the VIAME software to become more accurate at distinguishing fish across species, especially in dense and speciose videos, presents a promising pathway to achieve automated survey efficiency.

  • Topic Area
    Super Intelligence
    Study Type
    Evaluation > > Formative