Developing technology to recognise cod and salmon

By implementing deep learning (see fact box), scientists at the Institute of Marine Research (IMR), Scantrol Deep Vision, the University of Tromsø and the Norwegian Computing Centre will further develop existing image recognition and data analysis techniques. Previous studies have demonstrated that these methods are very reliable at distinguishing different species.

Using “intelligent” machines to save time

The amount of data collected by the IMR keeps growing. In 2017, more data was collected than in the whole of the previous history of the institute.

– Nearly all of our research and advice is very data-heavy. For example, we collect large amounts of data using echo sounders and trawl cameras. These data – or images – can provide important information about the fish stocks we are studying. We need to know the species, the length of the fish and other information that will eventually be incorporated into our recommendations on fishing quotas. We will save a lot of time if we can train machines to do some of this work for us, says Nils Olav Handegard, who together with Ketil Malde is responsible for the COGMAR project. The grant comes from the Research Council of Norway’s IKTPLUSS initiative.

Many applications

In a pilot project to COGMAR, scientists developed a technique for automatically classifying fish that achieved a recognition rate of up to 90 percent. There the species being identified were mackerel, herring and blue whiting.

– The technique can be extended to other marine species, and it can even – for instance – distinguish wild salmon from escaped farmed salmon, says Nils Olav Handegard.

– There are many areas of our work where we can see this technology being useful. It’s both about improving the quality of what we do at the moment and offering new services.

Modernising data processing

Nils Olav Handegard is delighted that the IMR was successful in its bid for ICT funding and pleased to be collaborating with the Norwegian Computing Centre.

– IMR possesses the large quantities of data required for deep learning. The Norwegian Computing Centre is naturally most interested in the technology itself, while we need it for specific applications. We are also in a position to start using the new techniques quite quickly. That was probably one of the main reasons why our application was successful, he says, adding that the COGMAR project will also play a role in modernising all data processing at the Institute of Marine Research.