Automated krill body length estimation based on stereo camera images

Mette M. Svantemann*, Bjørn A. Krafft, Fletcher F. Thompson, Guosong Zhang, Ludvig A. Krag

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

3 Downloads (Orbit)

Abstract

The fishery for Antarctic krill (Euphausia superba) is the largest by tonnage in the Southern Ocean, and understanding its population dynamics is essential for the sustainable management of this fishery. The standard method for calculating Antarctic krill biomass relies on hydroacoustic survey data and incorporates krill body length data collected concurrently. Traditional scientific acoustic surveys involve manually measuring the body lengths of individual krill caught using fine- meshed nets or trawls along acoustic transects. This work is resource-demanding and could represent a source of human error. To address these challenges, we develop and test an alternative, more automated method for estimating krill body length data by employing an in-trawl stereo camera system. This system collects images that are automatically processed by a custom-trained machine learning model. The results from the machine learning model are then compared to manually measured krill subsampled from the total catch of the corresponding trawl hauls. We demonstrated the ability to extract body lengths from underwater images. However, our results highlighted uncertainties, which we propose addressing by incorporating more advanced camera technology and optimizing the observation section of the small-meshed two-layer krill trawl.
Original languageEnglish
Article numberfsaf058
JournalICES Journal of Marine Science
Volume82
Issue number5
Number of pages11
ISSN1054-3139
DOIs
Publication statusPublished - 2025

Keywords

  • Feedback management (FBM)
  • Acoustic survey
  • Machine learning
  • Stereo imaging
  • Length estimation

Fingerprint

Dive into the research topics of 'Automated krill body length estimation based on stereo camera images'. Together they form a unique fingerprint.

Cite this