Machine learning involves letting advanced software automate work tasks and provide better information about conditions on the fish farm. Lice counting, estimating the weight of the fish, producing growth curves, and diagnosing the health of the fish are some of the areas of use, and there will be more.
1 What is machine learning?
Machine learning is a form of artificial intelligence. It is a software, a network of algorithms, that can be trained to recognise and interpret patterns in images and data. The technology behind it is the same as that used in face recognition on Facebook, interpreting the road for self-driving cars – or to recommend films and music for you on Spotify and Netflix based on previous choices.
2 What is the difference between regular software and machine learning software?
In a normal computer program, a calculation is performed on the basis of data that is entered into the program. In machine learning, the programs and algorithms evolve as they are "trained" with new sets of data. There is a form of pattern recognition, where the algorithms learn what to look for. They receive new training data together with corrections to the answers they have provided previously, and this enables them to produce increasingly precise results.
3 So how do you train the algorithms?
Let us assume that the algorithms have to be trained to recognise salmon lice. The algorithms are fed with training data, a lot of pictures of salmon with lice in different stages. By providing feedback to the algorithms, they become increasingly accurate in recognising salmon lice. It is important that the pictures are taken under real conditions in the cage, with changing light and different angles of the fish.
4 Do some algorithms learn more quickly than others?
The quality of the algorithms is important. They may need to be adjusted along the way to emphasise other factors in the data so that the results become more accurate. The number of algorithms through which the data moves might also need to be expanded. For example, the software might be expected to recognise the identity of individual fish based on the unique dot pattern each of them has. If the algorithms also have to recognise the same fish over time, they must have the capacity to understand how the dot pattern changes with the growth of the fish.
5 How to obtain the images for interpretation?
The pictures are taken with an underwater camera in the cage. Per Erik Hansen, Customer Success Manager at Aquabyte, explains that they use a standard camera with particularly good optics and two lenses in order to be able to measure the distance to the fish. The camera takes a continuous stream of images and is positioned so that as many fish as possible swim past. Software in the camera eliminates the images that cannot be used and sends the best ones for analysis.
6 Can you trust the results?
“Those of our customers who have used the system for a long time, report that the results are pretty much in line with reality", continues Per Erik Hansen. The fish farmers receive daily lice counts for each cage, with lice counting for far more fish than is possible when counting manually. “The software distinguishes between individual fish so that the same fish is not counted more than once. This contributes to Aquabyte delivering highly accurate lice numbers”, says Hansen.
7 What can you use it for?
Precise counting of salmon lice and accurate estimation of biomass and weight distribution are two of the areas of use that have fully developed systems. Fish health is another important area of application. Detection of winter lesions, deformations and other external changes on the salmon are relevant uses. By giving fish farmers much better insight and an improved decision-making base, it will be possible for them to make better decisions and increase the efficiency of fish farm operations.
Aquabyte has developed a system for counting lice using machine learning, based on images from an underwater camera with excellent optics. The system also delivers extremely good results for measuring biomass, using weight distribution and analytical tools.
Lice counting and biomass both use the same system with regard to the camera in the cage.