Thorvald showcases our prize-winning salmon trap at CatchID, Tromsø

On the 19th to 20th of March this year, Thorvald went to the International Symposium on Catch Identification Technologies (CatchID) in Tromsø, hosted by the Norwegian Directorate of Fisheries. He presented our innovative approach to addressing the challenge posed by the invasive humpback salmon in Norway. This non-native species has been outcompeting the indigenous Atlantic salmon, creating significant ecological and economic concerns.

The Humpback Salmon Issue

Humpback salmon, also known as zombie salmon due to their post-spawning decay, have been migrating up Norwegian rivers in alarming numbers. In 2023, around one million humpback salmon were estimated to have entered Norwegian rivers, a tenfold increase from previous migration seasons. Despite efforts to halt their progress, only about 250,000 were stopped, leaving a substantial threat to the native Atlantic salmon population.

The Technological Solution

To combat this issue, our team, in collaboration with Huawei, Troll Systems and the Berlevåg Hunters and Fishers Association (BJFF), developed an AI-driven solution. This system uses a combination of physical barriers and AI technology to distinguish between humpback salmon and Atlantic salmon.

The system consists of an underwater tunnel equipped with a camera and a gate. The AI vision system identifies the fish species as they pass through the tunnel. If an Atlantic salmon is detected, the gate opens, allowing it to continue upstream. Conversely, if a humpback salmon is identified, the gate remains closed, redirecting the fish into a containment area.

This solution ended up winning prizes at the AI festival in Cannes and the Global Mobile Awards 2024. Follow the links to read more.

Technical Challenges and Solutions

Thorvald highlighted several technical challenges addressed by our solution:

Real-Time Processing: The system processes high-bitrate video streams in real-time, making decisions within 0.2 seconds to open or close the gate.

Training Data and Model Generalization: Training data was sourced from different locations, posing challenges due to varying conditions. The team overcame this by developing robust models capable of generalizing across different environments.

MLOps Integration: The system includes machine learning operations (MLOps) for deployment, monitoring, and evaluation. This ensures the model functions effectively in real-world conditions and allows for continuous improvement through cloud-based analysis.

Physical and Environmental Considerations: The system operates in remote locations with poor internet connectivity and unstable power supplies. Solutions include reliable edge devices and efficient data transmission to the cloud for further processing.

Building Trust with Stakeholders

A critical aspect of the project is maintaining trust with local stakeholders and communities. The AI system's performance is continuously monitored, and any significant deviations from expected outcomes trigger human intervention. A web application provides transparency, allowing stakeholders to see real-time data and insights, further building trust and cooperation.

Future Engagements

Thorvald concluded by emphasizing Simula Consulting’s interest in expanding its involvement in the fisheries industry. He encouraged collaboration on both research and commercial projects, highlighting the potential for AI and machine learning to drive innovative solutions in this sector.

This presentation at the CatchID conference showcased a practical application of AI in addressing ecological challenges, demonstrating how technology can contribute to sustainable environmental management.

The talk

Watch the full talk in the video below.