STORY / PUBLIC SOURCE RECORD
Marine science: training artificial intelligence to identify fish
- Published 27 March 2023; updated 21 September 2026; estimated reading time 4 minutes. The article describes Institute of Marine Research (IMR) and Norwegian Computing Center research using AI to label acoustic echo-sounder images from North Sea sandeel surveys. Echo signatures help scientists identify species, estimate stocks and advise sustainable harvests. Researchers trained the model on five years of an 11-year survey series previously labelled by IMR technician Ronald Pedersen, then evaluated it on the other six years.
- Against Pedersen’s labels as the reference, the model scored 8.7/10 at distinguishing sandeel, empty water and other fish. IMR scientist Nils Olav Handegard says the aim was to share repetitive workload rather than replace researchers. The page recounts a 23 April 2020 survey where AI was to analyse data alongside the human technician; that is historical context, not a current survey update.
- IMR was also testing autonomous sailing drones for sandeel surveys. The article says onboard AI could interpret large raw datasets and send results in real time where raw data volumes exceed satellite transmission capacity. Planned next steps at publication included applying the approach to other species/surveys and exploring unsupervised learning.
- Reference given: Brautaset et al., “Acoustic classification in multifrequency echosounder data using deep convolutional neural networks,” ICES Journal of Marine Science (2020), DOI 10.1093/icesjms/fsz235. The page credits original reporting by Erlend A. Lorentzen, first published by IMR on 11 May 2020. Related industry: Technology. No direct contact block shown.