We all know that Buridan's ass starved in the end because it couldn’t decide between two equally large and equally distant piles of hay. This thousand year old parable that is said to have been written by the Persian philosopher Al-Ghazali effectively describes a dilemma we still often face today when we are asked to choose between two supposedly identical alternatives. In computer science we speak of a "deadlock" when the available alternatives for solving a problem virtually "jam" and block each other. All categorizing, grading, graduating and classifying, separating and sorting derives from detectable differences between products, clearly defined criteria which make this kind of allocation possible at all. The clearer the differences, the easier it is to make decisions and the more accurate the decisions will be. Fishes and seafood are natural products, which makes individual deviations from the "species-specific standard" virtually inevitable. From the recognition and removal of damaged fishes and unwanted catches to the quality-based classification of processed products, decision-making is required every day at all levels of the value chain.
The English language makes a fine difference between “sorting” und “grading”. Although there is some overlap with similarities between the two processes the term “sorting” is primarily used for processes involving separation and differentiation of seafood products according to objectively measurable criteria such as exact species, size, weight, maturity and sex. In contrast, "grading" is used more to classify and evaluate on the basis of complex decision criteria that can be distinguished in degrees and also be subjectively influenced. A typical example of grading would be the quality assessment of a fish or product that can be influenced by many factors such as freshness, colour, appearance, or overall impression. The transitions between the different assessments are often smooth and differences are then represented using a scale to denote varying grades or quality levels. The objective of both sorting and grading, however, is to establish a greater "uniformity" within defined product groups. Good sorting helps to avoid losses in the production process. A lot of machines operate more effectively and achieve a higher yield when processing more uniform raw materials. Standardized products are more attractive to customers and often have a higher market value. Farmed turbot in the 3-4 kg range costs 12.15 EUR / kg on average which is about 50% more than the price demanded for small fishes weighing 1-2 kg (8.50 EUR / kg). Vannamei shrimp (HOSO) in large counts of 20-30 per kg cost 8.96 EUR / kg which is about double the price of small 91-110 counts. In the case of seabass from aquaculture the price differences between the different sizes are even more pronounced: small fishes (200-300 grams) cost about 3.80 EUR / kg, particularly large specimens over 1 kg around 10 EUR / kg.
In general it is possible to sort and classify fish and fish products either manually or by machine. Manual sorting demands a keen eye, a quick hand, accurate product knowledge, and a lot of experience. The wider the range of different quality levels, the harder it will be to come to a decision. Where large amounts of products have to be classified it will hardly be possible to avoid mistakes. Although the use of sorting machines reduces the stress and workload for the staff there are not always suitable mechanical solutions for every need… especially since even the best technology does not relieve a person of the decision on the criteria according to which the machine should separate the goods. Great hopes currently rest on automated sorting systems that use methods of intelligent image recognition and powerful sensor technology. They capture not only a variety of conventional data such as size, body shape, flesh texture or colour but could also provide additional information on the freshness, water or fat content of the fillet. Such automated systems could thus enable sorting and grading to be carried out in a single process.
Unwanted by-catch already rejected in the net
During the fishing process, initial sorting sometimes already takes place in the fishing net. Sorting grids, escape windows and other selective devices are used to ensure that only the target species get into the net, and unwanted by-catches are avoided. Starting points for selective fishing are primarily the species-specific size and behavioral differences between the fishes which try to escape either upwards of downwards from the net, for example. The electronic "Fish Selector" from the Icelandic company Star-Oddi can allegedly even distinguish fish according to size and species underwater in the fishing net. The Fish Selector uses automatic image recognition to identify the target species and unwanted fishes are diverted out of the trawl. Since the device can be programmed not only to certain species, but also to their size it will also prevent catching juveniles. Apart from that, the Fish Selector collects and stores important additional information, such as fishing depth and water temperature which are of significance for fisheries management. Despite the advantages, however, it is questionable whether the device will be used widely in practice. It has to be installed at the net opening, which would probably change the hydrodynamic properties of the net. And in addition, if the net is “lost” or destroyed the expensive Selector would also be gone.
That is why many fishermen probably prefer to sort the catch on board for the time being as specified in EU regulation 3703/85. Every small shrimp boat in the North Sea for example, has a sorting drum on board to sieve the shrimps from the catch and return the by-catch to the sea. Carsoe Seafood Processing provides sorting systems which can be installed either on board or on land as single units or integrated into complete processing lines. The Bycatch Separator, for example, examines the body shapes and sizes of the fishes. It functions mainly on the basis of grid slits that gradually widen in the direction of product flow. Thus, small products fall through first, larger ones follow accordingly later. This allows a continuous sorting process without product jams. Current sorting systems separate the product flow into four to seven size categories, depending on the device’s construction.
Sorting technology based on image recognition still in its infancy
While demersal species such as cod and saithe are usually sorted by hand, mechanical solutions are often used for pelagic fish that occur in much larger quantities. The fishes slide individually down an inclined grid made up of gradually widening slits until they fall through the grid according to their size. Some sorting machines have fixed grids that vibrate strongly to move the fish to the desired position for sorting. Other systems use diverging rollers or conveyor belts between with grid slits in between. This simple technology has a high sorting capacity but also a relatively high error rate because it bases individual differentiation on the thickness of the fish. However, the thickness of a fish correlates only poorly with its length, which is why this method often leads to incorrect sorting. With the Smartline Grader from Marel such errors can, however, be eliminated. The system is flexible, accurate, and works fast. It can be used for sorting both whole fish and fillets and pieces by size.
In this area, too, many machine manufacturers are working on automatic systems which with the help of computerized image recognition can register fish species as well as their length and weight. There are already viable sorting systems available on the market that can recognize more than 20 species of fish, more than 100 colour variants, and a good dozen variable characteristics in body shape. However, many systems prove too slow under working conditions and the error rate is relatively high. It is particularly juveniles that present problems for they often lack the species-specific characteristics upon which differentiation is based. Flatfishes are also not easy: they look different on the eye and blind side. More powerful and more reliable systems are possible in principle but would often be so expensive that investment in them would hardly be worthwhile for the companies concerned. However, when it comes to complex assessment of subtle quality characteristics current image-based sorting systems come up against their performance limits, too. When classifying and sorting0020stockfish, which can include more than a dozen categories of “prime” and “second grades”, automatic systems will probably not be able to replace manual work and the human eye for the time being.
Sorting by size enables optimal feeding
Sorting requirements in aquaculture are particularly varied and wide. They range from the egg, through fry and fingerlings, to the adolescent and marketable fish. For some fish species, fry measuring only a few millimeters have to be separated according to size because cannibalism often occurs in this early phase of life: the larger fishes attack their smaller siblings. Fast and slow growers must also be separated regularly because sorting by size enables optimal feeding and more uniform growth of the fishes. Requirements are particularly high when sorting live fish because injuries and loss of scales must be avoided at all costs. Any intervention into the familiar living environment and situation is stressful for the animals and mostly leads to refusal to eat and poorer growth. For this reason they should remain in the water for as long as possible or at least be kept constantly wet. Mechanical intervention and direct hand contact should be reduced to the unavoidable minimum. Sorting fry is particularly challenging for they are often sensitive to external influences. If possible, one should use the natural activities of the fishes themselves for the sorting processes, for example allowing them to swim independently through grids with different gap widths. Such sorting scales and grids are simple, work reliably and are inexpensive, but only allow separation into two size groups. Small fishes slip through the slits and larger juveniles remain on the grid. A flexible solution for multiple sorting are sorting grids whose outer frame can be moved like a parallelogram. This reduces the width of the grid slits so that several size groups can be screened in succession with the same device.
The range of available sorting technology is extremely broad. In practice, one finds both universal solutions for various application fields as well as individual systems that have been specially developed for a particular user’s existing operational structures. Not every method and technique is equally appropriate for each species. It is particularly difficult, for example, to sort eels, which are very flexible and can change their body shape, which is why current sorting methods fail with them. In contrast, the grading and sorting of fish eggs, which usually occur in very large quantities would hardly be possible without mechanical assistance. The AGM Fish Egg Sorter from Skala Maskon assesses the quality of trout and salmon eggs for example on the basis of several characteristics by computer analysis. To do this, two photos are taken of each egg from opposite sides and evaluated using special software. Not only unfertilized and dead eggs are recognizable, but also other criteria such as twin, small-eyed, and coagulations in the fish embryo. Abnormal eggs can thus already be recognized at an early stage which reduces economic losses and the risk of diseases during the subsequent farming process.
When sorting larger fish automatic graders are often used such as those offered by Faivre and Milanese. These systems are very gentle on the fish because the fishes remain constantly in the water except for the brief sorting process. Using spiral conveyors, bucket conveyors or hydro-pneumatic fish pumps they are carried directly from the tank to the sorting machine where they pass over counter-rotating rollers between which the gap gradually widens in the direction of the product flow. During this process they are constantly washed over with water by powerful pumps to keep stress to a minimum. Depending on the fish size and the construction of the grader differentiations are possible into several size groups. In the case of juveniles and smolts up to 100 grams, the number of groups is usually between five and eight, with larger animals three or four. After the animals have passed between the rollers they can often swim through plastic pipes directly into the tank provided for their size group. Such sorting systems can be fitted with optional fish counters and weighing scales for recording and monitoring of fish stocks. Even more accurate results and more gentle treatment of the fish is promised by the use of advanced ultrasound technologies that some equipment developers are already testing. Automatic UltraGraders could reduce the sorting workload even further for they enable greater accuracy without endangering the fish. In the systems developed so far, which are mostly still in the experimental phase, the fishes are separated and guided to the sensor unit. There a photoelectric sensor detects the fish’s length while at the same time its body thickness is measured by ultrasound. From this data the computer can calculate the weight of each fish and assign it to the appropriate group.
Objective measurement or subjective perception?
What is needed in the fisheries and aquaculture sector is also required in the processing area. At many points in the production process products have to be sorted and evaluated in order to reduce the variability in sizes, shapes and weights, in textures and colours. At the slaughter lines for salmon this often begins with image recognition systems that enable accurate, precise contour cuts when heading, filleting and trimming the fish. Minor faults, such as defects in the scales, short-tailed or hump-backed, are recognized and taken into account during category allocation of the products. Optimization of the cuts reduces losses, provides more consistent results and thus enables better quality. So far, the quality of a salmon is usually classified visually and manually. This method works well, but is not completely free of errors. Image recognition systems could objectify selection using measurable parameters and show deviations from the species-specific standard objectively and unerringly.
Basically two important goals are to be distinguished during classification and evaluation in the production sector. "Defect grading" is primarily used to eliminate defective products from the production process, for example because the specified weight is not correct, due to the discovery of foreign bodies in the packaging, or because additives were declared incorrectly. "Value grading" on the other hand arranges the products into specific individual groups on the basis of certain quality criteria, consumer acceptance and the products’ commercial value. It may be economically worthwhile, for example, to take absolutely flawless products as desired by some consumers out of the regular range and market them in the premium segment. Image recognition systems are probably not capable of such selection procedures, however. They are only suitable for "hard grading" which evaluates products according to objective, strictly defined criteria. "Soft grading" (also called fuzzy grading) focuses on the subjective, emotionally influenced perception of the products, on the external impression, appearance, harmonious shapes and sensory properties such as firmness, colour and smell. It will probably still be some time before machines will be able to judge and select according to such criteria but it cannot be fully ruled out. Cabinplant already offers sorting machines that evaluate salmon fillets on the intensity of the red colour of their flesh in accordance with industry standard NS 940 to enable producers to serve specific market segments according to their accurately defined colour preferences.