Machine Vision Can Help Sort Fish

Consumers prefer fresh fish, with an even colour and shape. Salmon fillets are sorted by hand, with employees grading them on their shape, colour and the presence of any surface injuries. Grading has been difficult to mechanize, as a result of the variety of factors that influence the fillet’s quality. For example, remnants of blood in the stomach cavity and the level of stress the fish experienced before its death can lower quality.

A machine vision and image analysis technique would enable rapid, accurate sorting of bad salmon fillets (above) from good ones (below).

A researcher at Sintef Fisheries and Aquaculture Research, Ekrem Misimi has created mathematical algorithms that could enable machine vision sorting of fish, standardizing and simplifying the process of determining quality. The algorithms, part of the investigator’s doctoral thesis, combine machine vision with pattern recognition to input geometrical descriptions of the size, colour and shape of salmon into a computer.

Whereas sorting fillets by hand requires workers to compare the colour of the fish to a colour-matching card, the new method takes photos of the cards and stores the values obtained from them to compare the colour value for the fillet being graded with the stored data. Misimi said that the image analysis technique would sort the fish into “production,” “ordinary” and “superior” classes with 90 per cent accuracy and would not require physical contact with the fish.