Computer vision protects endangered fish in Norway

In Norway’s rivers, an endangered species of salmon is pulled from the brink of extinction by AI

In some Norwegian rivers, an invasive species of salmon was crowding out a domestic variety. What role did your company play in solving this problem?
In the 1990s, humpback salmons were artificially introduced into Norwegian rivers. Over time, because of climate change, conditions became much more advantageous for the humpback salmon, allowing them to take over the rivers more and more. They compete for resources with the Atlantic salmon, which is the native variety.

We helped build an AI computer vision model that can distinguish between the two. 

How does it do that? 
It works as an image classifier. The ones you see on the internet can distinguish between, for instance, a cat and a dog, right? That's because they have been trained on lots of images of cats and dogs. 

Essentially, this works exactly the same. It has seen tens of thousands of images of different species of fish: the Atlantic salmon and the humpback salmon – but also arctic char and smaller fish. 

We worked with other partners in the project to build entire solutions, but our focus has been on the AI and the software backend. We trained the computer vision model and set up a data-processing pipeline, making sure that all the images made it from the camera to the server and could then be sent to the cloud. 

What happened to the humpback salmon as a result?
The invasive humpback salmon gets separated from the Atlantic salmon. The native salmon are allowed to swim freely in the river, while the invasive species is shunted into an underwater cage, so it’s no longer a threat. 

All of the data that's collected is sent to river managers, enabling them to see not just how this fish trap is doing, but also, how many fish they have seen. Is that different from other years or from last week? In essence, we can use the data to evaluate the river’s overall health.