A recent scientific contribution by researchers from DTU Aqua, an OptiFish project partner contributing to Electronic Monitoring and catch data research, was presented at the Northern Lights Deep Learning (NLDL) Conference 2026 in Tromsø, Norway (6–8 January 2026), with reference to the OptiFish project.
From manual video review to automated intelligence
Electronic Monitoring (EM) systems are increasingly used in commercial fisheries. However, a significant bottleneck remains the manual review of large volumes of video data, which is time-consuming and resource-intensive.
To address this challenge, the DTU Aqua research team presented their work titled: “Towards Visual Re-Identification of Fish using Fine-Grained Classification for Electronic Monitoring in Fisheries.”
The research focuses on automating EM processes by applying metric learning techniques for the visual re-identification of fish in commercial fisheries.
Why fish re-identification is challenging
Onboard sorting belts are highly dynamic environments. Fish are constantly moving, overlapping, and becoming partially occluded, conditions that often cause standard tracking models to lose individual identities across video frames or camera views.
The proposed approach applies metric learning-based feature similarity measures to enable the maintenance of individual fish identities even under these challenging conditions.
Evaluation on benchmark data
The method was evaluated using the AutoFish benchmark dataset, developed by the Visual Analysis and Perception Lab of Aalborg University. The results demonstrated the feasibility of:
-
Maintaining individual fish identities across consecutive video frames
-
Re-identifying fish across different camera views
These results indicate that the approach is a key step toward fully automated catch composition analysis.
Relevance for fisheries monitoring
Automated visual re-identification of fish can contribute to improved data accuracy for stock assessments and support sustainable fisheries management by reducing reliance on manual video review.
Connection to the OptiFish project
The OptiFish project was referenced during the poster presentation, and the project’s grant agreement is acknowledged in the paper. The research topic is related to OptiFish activities on Electronic Monitoring and AI-based visual analysis for fisheries data collection.
Authors
Samitha Nuwan Thilakarathna (Corresponding author | )
Ercan Avsar
Malte Pedersen
Martin Mathias Nielsen
This research was conducted within the Section of Fisheries Technology, DTU Aqua.
Read more
- Read the full publication in OpenReview or arXiv
- Conference proceedings (NLDL 2026)
Words by the reframe.food team.