AI streamlines the work of rebuilding fish stocks

The background for the project was Norconsult's assignment to rehabilitate an old fish ladder at Laksforsen in order to rebuild the salmon population in Vefsna. Part of the work involved reviewing video footage and manually counting fish that went through the stairs.

The machine learning environment in Norconsult Digital was tasked with streamlining and automating the time-consuming process. They created a machine learning model that counted the fish population by detecting fish in the video material. The model was based on several hundred hours of video material from the fish ladder.

Started by quality assuring the data material

Victoria Glott, part of the team at Norconsult Digital, worked in particular with the pre-processing of the dataset to create the best possible basis for training the model.

She explains the procedure: "We started by inspecting the dataset and assessing the quality of the videos. The starting point is always that the data must be of high enough quality. For example, lighting conditions affected the visibility of the video, and sometimes it was too cloudy".

After weeding out these issues and spending time on data processing, they considered different approaches to training the model on the video material. They noticed what was fish in the videos and used this as a data basis in the machine learning model.

"After the machine learning model had been trained, we were able to test the model and evaluate whether it was sufficient. We compared the figures from the model with manual counts, and the results were very good. Considering the amount of data we had, this gave a satisfactory result," says Victoria.

Created something that can be used for other animal counts

The efficiency and automation that AI offers is helping to revolutionize how we can approach and solve environmental challenges, not only in aquatic environments, but also in other ecosystems where similar challenges exist. The experience from Laksforsen can serve as a model for other environmental monitoring projects.

"In another project, we have used similar technology for recognizing animals, for example on wildlife cameras in wildlife crossings, so we can see the ripple effects of it," says Victoria.

Norconsult Digital offers end-to-end solutions, from business understanding, data processing and analysis, model and user interface development, and production setting adapted to the customer's needs.