Artificial intelligence (AI) is a hot-button issue these days.

For every big business pushing AI there are thousands of concerned voices raising alarms over applications such as image generation and AI-powered surveillance tech, or the inability of school children to complete their work without ChatGPT.

But, there are many non-controversial and ethical AI applications which have served to open new opportunities for businesses, including in the fish farming sector. Perhaps if these applications received more attention, AI’s reputation would improve.

Hatchery International spoke with a variety of businesses and researchers who offer AI-solutions for early fish rearing to find out more.

Counting & tracking

One of the most common AI applications for fish farms and hatcheries is in fish counting. Innovasea offer HydroAI, which they say replaces ‘the labour-intensive, error-prone process of manual counting with automated real-time fish counts.’ Vice-President of Research and Development Jean Quirion says that the impetus for the product came from a customer who wanted to track fish around hydropower facilities.

“We thought we could solve this problem differently than our traditional fish tracking methodologies by using a camera that can see fish and an AI and computer vision to automatically detect these fish and monitor them,” says Quirion. “So we explored this idea with this customer as a partnership and we realized that it is a market trend that’s bigger than this one customer, so we set out to develop HydroAI as a solution for tracking fish around hydro.”

While HydroAI was developed with tracking fish around hydropower facilities in mind, it is appropriate for tracking fish in any sort of narrow passages – those just happen to be plentiful around hydropower facilities. The system requires a camera to be positioned in such a way that it can see the fish that are moving by and a narrow enough passage that no fish are missed by the camera – Innovasea will provide a service to ensure that the camera is installed properly and offer a custom insert that can attach to the passage for easier mounting logistics.

The AI models have been trained to detect when a fish is in the camera view and also to track fish individually that move through the camera frame. Generally, in these applications, fish are only moving in one direction so there’s no concerns about accidentally counting fish more than once. If a fish enters the camera frame, changes its mind and turns around that counts as a ‘zero.’

“In other words, we can tell if the fish is moving upstream or downstream and we can account for the net number of fish that the system has seen,” says Quirion. “Even though there might be 10s of fish in the camera frame, the model can tell them apart as individuals and not double count from frame to frame.”

Quirion says that the next phrase they are working on is species classification. They have developed the basic capability and expect to have it ready to roll out publicly in 2027.

Hatchery International

Message History

Delivery option(s)