Fraunhofer Institute for Computer Graphics Research IGD
The OTC-SmartFishing project aims to make fishing more sustainable and efficient with the help of AI and state-of-the-art underwater technology. The focus is on developing a sensor system that detects and classifies fish directly in the net. The solution addresses the socially relevant issue of protecting marine resources and contributes to the sustainable use of the oceans.
Within the OTC Rostock innovation cluster, OTC-SmartFishing developed an AI-supported underwater camera system for the fishing industry. The prototype tested in the project detects and tracks fish in the trawl net. Together with the project partners, extensive annotated datasets could be created. OTC-SmartFishing provides a crucial step forward for ecologically selective fishing and for faster transfer into research and practice.
Project Description
Objectives
The goal of the project is to develop an optical underwater sensor system for in situ detection and classification of marine organisms for fishing and research. The core idea: a square-shaped net tunnel with a camera and LED light installed in one corner, covering the entire cross-section of the net and enabling—for the first time—quantitative, selective detection. This is intended to reduce unwanted bycatch and automate and improve data collection in research.
Work by Fraunhofer IGD
Together with the Thünen Institute of Baltic Sea Fisheries, IGD was responsible for the central software pipeline for machine learning. IGD structured the annotation workflow, built domain-specific datasets in close cooperation with project partners, and developed AI functions for object detection, tracking, and counting of fish. It evaluated learning approaches, defined quality metrics, and integrated the results into a real-time capable prototype.
Results and Outlook
The prototype delivers reliable detection results. The evaluation metrics confirm the general suitability of the methods but also highlight limitations caused by water turbidity, occlusions, and fish “flipping.” Next steps include expanding and balancing the datasets as well as improving re-identification and movement models. This lays the foundation for more selective fishing, faster stock assessments, and further development within the OTC innovation cluster.
