Modern aquaculture has a major challenge in observing both the behaviour and the condition of fish in a production setting in sea pens. It is of paramount importance to be able to observe the fish to estimate size, and fish health. There are numerous companies trying to solve this obstacle through different solutions, popular methods include cameras, and frames that register fish as they swim through it. These methods have in common that they register individual (or a few individual) fish, and that they have limited range for observing the fish. This leads to long observation times before information on a population level can be obtained. Also, the images and or other sensor data needs to be interpreted and analysed before the data can be put together into a representation of the fish in the pen.
Through our project we aim to develop a methodology for observing the entire population in short measurement cycles or in near real time. We wish to use principles from hydro acoustics, but with innovative solutions that adapts the technology to the environmental conditions specific to Atlantic salmon.
We believe that the advantage of hydro acoustics is the ability to observe multiple individuals in a short time frame, and at the same time being independent of daylight. Further, hydro acoustics has the potential of observing the pen and the surrounding area in addition to the fish.
Together with our collaborators, NTNU Ålesund, we wish to use hydro acoustics to observe size distribution and to observe fish migratory patterns in a pen over time. Out research will focus on both size dependent measures, and how different parameters affects fish behaviour in aquaculture.
Current technology for analysis of biomass is largely base on spotlight measurement of individuals, without taking the natural variation and biomass dynamics into the equation. This means that the fish needs to swim past a sensor, and the profiles is generated from a time series of this measurement.
The goal with this project is to develop a technology to provide analysis based on real time measurements at population level. With this technology one should be able to better understand the real time dynamic of the biomass, and thus gain insight in fish weight, behaviour in and surrounding the pen, appetite, welfare, health and maturation.
To exploit existing knowledge both from fish farming companies, NTNU and Furuno Norway we aim at building causal diagrams to better establish statistical models to explain variation in data. Through causal inference with multiple covariates we aim to build prediction-models based on new insight from datasets not available with current technology, combined with extensive domain knowledge both in the domain of fish biometrics (Furuno) and the biology of atlantic salmon (NTNU)
To develop this technology we aim at utilizing statistical multi model inference, causal diagrams and exploratory data analysis techniques combined with machine learning. To be able to develop this technology a better understanding of biological parameters is crucial for the accuracy of the product. Research is therefore needed to develop biological models taking into consideration the fish behavior and its biological variation. This research is intended to be done in close cooperation with NTNU.