Back to search

HAVBRUK2-Stort program for havbruksforskning

NordForsk - Intelligent farming and health control in landbased recirculating aquaculture systems

Awarded: NOK 2.6 mill.

Production of fish in land-based aquaculture contributes significantly to the production of animal protein in the human food-supply chain in the Nordic countries. Recirculation aquaculture systems (RAS) provide an environmentally sustainable solution that is becoming more and more relevant for current production systems. However, considerable challenges with these production systems means narrow production margins and uncertainties regarding fish health and production performance. An extensive amount of information on water quality, feed use and health parameters are gathered through monitoring of the fish and their environment via sensors. In most cases, these data are stored in separate datasystems, not utilizing the huge potential for more precise monitoring and reporting of the realtime health, welfare and growth of the fish. We aim to integrate such data from many farms and production cycles and through the usage of Artificial Intelligence (AI), enabling the farmers to move from experience-based to knowledge-based decision making. This will enable a more sustainable production with regards to production margins, environmental impact and fish health and -welfare. Specific objectives of the project are: * Integrating several sources of bigdata including data on water quality, feed usage, health and production * Developing a deep learning-based prototype system for audiovisual monitoring of fish behaviour in RAS farms * Development of a dynamic growth model that include multiple data sources * Implementation of precision fish farming system for optimal production * Development of a health monitoring system. The best indicator of health and well-being is the fish's own behavior. In the project, deep learning tools are used on video sequences of fish, i.e. teach computer systems how the fish react in different situations. In two RAS facilities, the fish's surface activity has been filmed, and a selection of video sequences has subsequently been annotated, i.e. people have marked what the fish's head is in the videos. These annotations have been used to train the system to recognize the fish, and be able to follow its movements in the tank. In this way, we have created an algorithm for normal behaviour. In the last part of the project, feeding experiments are carried out, so that the system can be trained to recognize feeding behaviour. In order to find out how the fish are affected by the production system, extensive welfare and disease investigations have been carried out on the fish in the two facilities. Similar investigations have been carried out at two other RAS facilities. During 2021-2023, we have followed 18 fish groups at the four facilities. Each fish group has been examined at least once, and in total there have been 41 examinations of a total of 2,828 fish. For the survey, we have used the system for scoring external welfare indicators from the Fishwell handbook. We found, that 98% of all fish have fin damage, and that these damages change over time, both in severity and which fins are most damaged. The results from the surveys have not been finalized yet, but will be published at the beginning of 2024. Production data, and data on feed consumption and environmental monitoring are also collected from all four RAS-farms. With these data, we will examine whether any health and welfare challenges are reflected in the collected data. In addition, production data from 15 RAS farms has been collected, and these are now being used to develop a model that should be able to detect deviations from normal or expected mortality, based on daily data. This data collection has also provided an insight into what types of data actually exist in RAS production in the Nordic countries. And also which types of measurements are or will be automated. In this way, we will get an overview of the status quo for the use of "big data" in RAS production, and be able to make suggestions for improvements.

Production of fish in land-based aquaculture contributes significantly to the production of animal protein in the human food-supply chain in the Nordic countries. Recirculation aquaculture systems (RAS) provide an environmentally sustainable solution that is becoming more and more relevant for current production systems. However, considerable challenges with these production systems means narrow production margins and uncertainties regarding fish health and production performance. An extensive amount of information on water quality, feed use and health parameters are gathered through monitoring of the fish and their environment via sensors. In most cases, these data are stored in separate datasystems, not utilizing the huge potential for more precise monitoring and reporting of the realtime health, welfare and growth of the fish. We aim to integrate such data from many farms and production cycles and through the usage of Artificial Intelligence (AI), enabling the farmers to move from experience-based to knowledge-based decision making. This will enable a more sustainable production with regards to production margins, environmental impact and fish health and -welfare. The best indicator of health and performance is the behaviour of the fish themselves. One major challenge in RAS is waste of feed. We will use AI and statistical models to develop a closed loop control of feeding to reduce feed waste and optimize growth, moving from a manual analysis and response to a knowledge-based automatic one. This will be done by the application of deep learning tools on video sequences of fish, teaching computer systems how the fish react when they are hungry and when they have been feed. In addition, statistical models will be used on the integrated datasources and video sequences to develop a system for early detection and warning of upcoming disease events, thus giving the farmer time to react in the narrow time-window between an adverse health event and death.

Funding scheme:

HAVBRUK2-Stort program for havbruksforskning