The fish-processing industry grades salmon manually into various quality classes. If sensors took over this job, costs could be cut and Norwegian companies could avoid having to set up shop abroad.
Ekrem Misimi, a research scientist at SINTEF Fisheries and Aquaculture Research, has recently defended his doctoral thesis on accurate mathematical descriptions that enable machines to sort fish according to quality.
Misimi has combined machine vision with pattern recognition methods, and has fed geometrical descriptions of the size, colour and shape of salmon into a PC, which then grades the fish according to its quality.
“The Norwegian fish-processing industry has been slow to introduce modern technology, and the production costs of a kilo of salmon in this country are an average of 5 – 10 kroner higher than in countries that compete with us. Exports of processed salmon are also still low, so the industry has a lot to gain by adopting these new methods,” says Misimi.
Today, fish are graded manually by employees who assess their shape, colour and any surface injuries, since consumers demand salmon fillets that are fresh and regular in colour and shape. This can be difficult to achieve using current technology.
If the salmon was stressed at the moment of its death, it stiffens more rapidly, and when it is stored on ice its fillets change colour and shape faster than fillets taken from an unstressed fish. Stressed fillets cannot be processed until they have passed through the stage of rigor mortis after two or three days, and meanwhile the product is losing freshness.
Moreover, there may be remains of blood in the stomach cavity from when the salmon was bled. This may leave flecks of blood on fresh and smoked fillets, a common cause of downgrading.
Colour is an important indicator of the quality of salmon fillets, and at present, a special ruler and a colour-matching card are used to sort the fillets that fall within approved limits from those that have to be rejected.
The new method simply takes photos of the colour cards and stores the values obtained, so that the colour of a fillet can be compared with values from the table. This objective method agrees well with the methods that human being use to analyse colours, and is also rapid and does not require physical contact with the fish.
“Machine vision and image analysis will enable us to sort fish into “production”, “ordinary” and “superior” classes, while revealing blood in the stomach cavity, with an accuracy of 90 percent. Automation can increase productivity and raise processing rates, while companies can avoid having to establish subsidiaries abroad,” says Misimi Ekrem.