Note: Descriptions are shown in the official language in which they were submitted.
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
1
METHOD OF CALCULATING QUALITY PARAMETERS OF FOODSTUFFS
The invention relates to a method for calculating quality
parameters of foodstuffs, particularly meat products,
especially. from fish. More specifically, the invention
relates to the use of a portable instrument in combination
with multivariate modelling for calculating colour, chemical
contents of pigment, fat and water in the field, for example
in a production facility, in a slaughterhouse or in a works
laboratory.
In what follows, reference is made to the calculation of a
number of quality parameters of fish, and to a particular
degree salmon, but the invention is not limited to the use on
fish, as it is conceivable for the method to be used also for
other foodstuffs, particularly other sorts of meat, and
particularly foodstuffs in which colour is a quality
criterion which has to be evaluated quickly and reliably.
By the characteristic size of an individual is meant one or
more dimensions, for example length,or diameter, area, volume
and/or weight, sufficient to provide a characteristic of the
individual. For a fish the characteristic quantity may be
described for example by length and/or.weight.
The measurements are made on an individual of the foodstuff,
.for example a slaughtered fish, and is based on the
measurement of reflected, visible light in combination with
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
2
information about the physical properties of the individual,
such as the length and weight of'the fish, and information
about date of sampling. Multivariate models are made,
calibrated against reference values.
A characteristic feature of salmonoids is that the muscles
have a distinct red colour. The red colour is caused
essentially by the natural pigment astaxanthin. Astaxanthin
is produced in phytoplankton and finds it way up through the
food chain through small crustaceans, which are then eaten by
salmonoids. There are also other pigments, such as
canthaxanthin, lutein, zeaxanthin, and in what follows, the
pigments are also called, by a collective term, carotenoids.
The degree of red colouring is considered by many consumers
as an important quality parameter when they are to buy salmon
in the form of fillets or chops. Salmon is also used by the
processing industry for producing, for example, smoked salmon
and gravlax .(brine-cured salmon). An important quality
parameter of smoked salmon and gravlax is the degree of red
colouring after processing.
As the red colour is important to the evaluation of the
quality of the products that reach the consumer, measuring
the red colour before removing fish for slaughtering is also
important. If the colouring of the fish is too poor, this
will give a deduction in the price to the farmer and the fish
will be difficult to sell, in particular in markets expecting
a bright red colour in the fish flesh.
To a certain degree, there is a connection between the amount
of carotenoids in the fish feed, such as astaxanthin, and how
much carotenoids is found in the muscles of the fish.
Different strategies have also been worked out for how the
farmer should most efficiently colour the fish. Thus, there
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
3
is talk about-colouring periods, in which the astaxanthin
content, for example, of the muscles is to be increased,
measured as mg of astaxanthin per gram of muscle, and
maintenance colouring, in which. the level of astaxanthin is
only to be kept stable as the fish grows. It is also known
that the absorption of carotenoids varies with fish size,
season and that there are hereditary differences between the
different releases of fish. Known in particular is the so-
called spring drop, in which the astaxanthin content in the
fish muscle decreases in spring. These different factors make
it necessary for the farmer to sample the fish stock
regularly to know the state of the fish as regards the
carotenoid content and possibly adjust the amount of
carotenoid in the feed in order for the colouring to be in
accordance with the production plan.
An alternative to astaxanthin has been the use of the pigment
canthaxanthin. This gives a somewhat yellower fish flesh than
the use of the pigment astaxanthin. The two pigments have
also been used together in different proportions of mixture.
From Norwegian patent No. 306652 is known that feeding feeds
with elevated contents of the amino acid lysine for a shorter
period before slaughter gives a fish muscle with a visually
brighter red colour without the astaxanthin content being
elevated.
The degree of red colouring can be determined visually by
comparing the red colour of the flesh to the red colour of a
standardized set of colour cards. The set of colour cards is
a collection of coloured cardboard cards, in which red colour
saturation and intensity increase from card to card. Best
known is the so-called Roche colour fan (Roche SalmoFanTM) in
which the cards are numbered from 20 to 34. The colour of the
fish flesh is given a Roche colour card value based on which
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
4
card is the closest to the colour of the fish flesh. This
test is basically easy to perform on completely fresh fish
and can be carried out without any other means than the
colour cards. The use of standardized colour cards is well
established, and the Roche colour card value is established
as a standard. A further description of the basis for working
out such colour cards are given by Skrede, G., Risvik, E.,
Huber, M., Enersen, G. and Blumlein, L.; Developing a Color
Card for raw Flesh of Astaxanthin-fed Salmon. 1990. Journal
of Food Science,, 55, 361-363.
The use of colour cards is a subjective way of determining
colour, in which several factors affect the result. Natural
light will vary with time of day and the meteorological
conditions. Attempts have been made to remedy this by the use
of a standardized illumination box (for instance the Salmon
Colour Box, Skretting AS, Stavanger, Norway in which the
light source is fluorescent tubes). A known disadvantage of
fluorescent tubes is that the colour content of the light may
vary over the life span of the fluorescent tube. Further,
such an illumination box has an open side, where lateral
light will enter to fall on the sample.
A piece of fish, such as a fillet, does not have an evenly
coloured surface. The fillet is built of muscle fibres
alternating between connective tissue and fat. This
.complicates the visual colour measuring.
It is known that a fish with relatively much fat in the
muscle looks paler and scores lower than a relatively lean
fish.
In addition, the perception of colour is different from
person to person. It has turned out in practice that
different observers may grade the same piece of fish with a
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
difference of 3 on the Roche colour card scale, for example
from 22 to 24. In addition it is of importance whether the
observer has an economic interest in the result. Thus, a
seller will have a tendency to grade higher than a buyer.
There has been an attempt to remedy the problem with using
colour cards and a standardized illumination box with a more
elaborate illumination box as disclosed in Norwegian patent
NO 317714. In addition NO 317714 discloses a method of
predicting the chemical contents of astaxanthin, fat and
Roche colour value by means of a photometric technique and
measurement of RGB colour values (R = red, G = green, B =
blue). The camera is digital. The drawback of this method is
that it requires a large and elaborate illumination box which
is not suited for being moved, but only for use indoors.
There are also strict requirements as to the quality of the
camera optics and mechanical components like aperture and
shutter. The patent owner also points out the fact that it is
important that the so-called CCD chip ("Charge-Coupled
Device") is kept at a stable temperature during exposure.
Another drawback is that the intensity of the light changes
over time, which necessitates routines for following up and
control.
The red colour can also be determined by analysing the
chemical contents of astaxanthin in the fish flesh. Chemical
analysis of astaxanthin is complicated and can only be
carried out in laboratories by trained personnel. Equipment
like HPLC ("High Performance Liquid Chromatography") is also
required for the analysis to be carried out. Thus, a fish
farmer must send away the fish or piece of fish to be
analysed. The answer will come after several days, and there
is a considerable cost in having'such an analysis carried
out.
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
6
Another method is the use of NIR ("Near-Infrared
Reflectance"). This technique is based on a reference
material, in which astaxanthin and other carotenoids are
determined by conventional chemical analysis. An NIR spectrum
is taken of the same material, and by means of multivariate,
statistical techniques connections are found between=spectrum
and chemical contents. This connection is expressed in an NIR
equation. This NIR equation is used to predict the contents
of carotenoids, for example astaxanthin, when new samples are
analysed.
An NIR analysis is quicker and less expensive than a chemical
analysis. The instrument itself is an expensive and
stationary instrument and it is not remunerative for the
individual fish farmer to invest in an instrument of his own.
Therefore, also in this case, the fish farmer must send the
sample away, and it takes several days before the answer is
available.
Measuring the chemical contents of, among other things,
astaxanthin in the fish flesh by means of an NIR/VIS
instrument (an NIR instrument also measuring visible light)
is used as the established technique for example by the
present applicant and by others.
In the patent document US 6,649,412 is disclosed a slaughter
line for fish, in which an NIR probe is placed behind the
tool that removes the intestines from and cleans the belly of
the gutted fish. The probe illuminates the fish muscle from
the open belly and provides. information about the red colour,
so that fish may be sorted according to the red colour.
NIR instruments come in many sizes and designs, and they are
used in calibrations against parameters corresponding to
those used by the applicant. Potentially, small portable NIR-
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
7
VIS instruments may also be used for the purpose. These use a
relatively wide wave range, which makes the calibration
relatively robust. But the instruments are generally
expensive, they will be problematic to standardize in such a
way that the same calibration may be used on the entire
instrument park, and the data processing is complicated as
many variables (signals or reflections from different wave
lengths) are used.
The problem with other known instrumentation is that the
equipment is too heavy and voluminous to be transported along
easily for field analyses. At the same time the
instrumentation is so expensive that it is not very relevant
for purchase for the individual farming facility. Recently,
smaller portable NIR instruments have been developed. These
overcome the problem of stationary instruments. But it still
remains that the instruments are expensive and that they have
to be calibrated at regular intervals.
Thus, in general, for the analysis of chemical colour and fat
content the fish must be sent to laboratories or analysing
instruments that are located centrally. This takes time and
makes it necessary for decisions to be postponed in
anticipation of the objective analysis results becoming
available. The postponement may result in the time space at
disposal for influencing the product quality being limited
unnecessarily.
The background of the invention is that there is a need for
quick and objective decisions on the quality parameters of
fish. Today colour can be determined manually (also in the
field) by means of colour cards under standardized light
conditions, but the method is subjective and relatively
inaccurate. The method should be so simple that it can be
carried out at the production facility, that is to say the
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
8
fish enclosures, or in any case in close connection thereto,
such as on feeding barges, work boats, piers or indoors on
land.
A colorimeter is used for colour measuring of surfaces and
can be used to determine the position of a colour in the so-
called NCS system. Attempts have been made to use
colorimeters to determine the colour of fish flesh. The
colorimeter primarily gives colour values expressed as XYZ
values, again expressed as L*, a* and b* values, L* being the
lightness factor (black/white), a* being red/green
chromaticity and b* yellow/blue chromaticity. Secondarily the
colorimeter gives the values "Chroma" (C*ab) and "Hue" (H ab)
These values. are functions of L*, a* and b* and are measures
of colour intensity and colour composition respectively. The
functions are
C*ab = (a*2 + b*2) 1/2
H ab = tan-1(b* /a* )
The colorimeter may be a hand-held instrument and is placed
to lie or stand on the sample in such a manner that light
does not enter from the side onto the surface to be measured.
The colorimeter has an internal light source and this is
calibrated continuously by software which is an integrated
part of the instrument.
Reference methodology used today is based on visual
evaluation of colour which is related to colour cards (for
example Roche SalmoFanTm).
The use of a colorimeter to determine colour is established
knowledge; for example, a surface may be measured, and then
the colour in the NCS system which is the best match may be
found. Within the fish farming trade colorimeters are used
for colour measuring in fish. Then the values from either L*,
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
9
a* or b* are used, and an attempt is made to correlate these
to colour cards. The correlation is then normally relatively
weak.
Even though the colorimeter indicates colour in several ways,
it has turned out to be difficult to find a good correlation
between the measured values and the chemical contents of
carotenoids, for example astaxanthin, and the visual colour
card value. This is connected to, among other things, the
fact that there are other natural pigments than astaxanthin
in the salmon muscle. Yellowish pigments may also be present,
such as lutein and zeaxanthin, pigments coming from maize,
for example. This affects the reading of the colorimeter. The
occurrence of canthaxanthin will also affect the reading of
the colorimeter. Christiansen et al. found that a colorimeter
was suitable for quantifying the astaxanthin content in
relatively pale fish (2-4 mg of astaxanthin per kg.), but
that the instrument could not distinguish between the
astaxanthin content in samples having higher astaxanthin
content. The same authors found that the use of a colour card
fan predicted the chemical colour content better, but this
prediction was not satisfactory either. (Christiansen, R.,
Struksnaes, G., Estermann, R., Torrissen, O.J. 1995:
Assessment of flesh colour in Atlantic salmon, Salmo salar L.
Aquaculture Research, 26,'311-321)
The colorimeter measures within visible light. Thus, it will
respond relatively little to the amount of fat in the sample:
This makes it difficult to achieve a correlation with a
visually read colour card value.
The use of only the measured value from a colorimeter gives
too poor a correlation with a colour card value and is
unusable in practice. At the same time it is impossible for a
user to understand how all three values together correlate
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
with a colour card value. For this purpose multivariate data
processing is needed.
Measuring in visible light alone describes to a somewhat
limited degree the chemical properties of a sample, and even
colour will be described better by incorporating other
parameters into calibrations.
The invention has as its object to remedy or reduce at least
one of the drawbacks of the prior art.
The object is achieved through-features described in the
description below and in the claims-that follow.
The object of the invention is to perform an objective and
quick analysis of the flesh quality of a fish, with an
emphasis on the chemical contents of carotenoids like
astaxanthin and canthaxanthin, chemical content of fat and
colour card value of the fish muscle.
It is a further object that this may be done in the field,
preferably way out on the floating parts of a fish farming
facility, like floating walkways and feeding barges. It is a
further object that the analysis should be reasonable,
thereby allowing it to be carried out repeatedly, so that the
farmer will have some help in planning his fish production as
regards the colouring of the flesh of the fish.
It is a further object that by providing an objective and
reliable colour measurement which is at least just as
reliable as the manual colour measurement using colour cards,
such apparatus-based colour measurement could serve as
objective documentation in the buying and selling of the
fish.
CA 02642482 2010-12-21
11
The object of the invention is thereby to carry out objec-
tive, quick analyses of the flesh quality of the fish by
bringing the instrument to where the fish are, thereby saving
time and costs of sending fish for sampling.
More specifically, the invention relates to a method of cal-
culating a range of quality parameters, for example colour,
pigment content, fat and/or water content, of a foodstuff, in
particular a foodstuff of animal origin, for example meat
from fish and mammals, characterized in that the method com-
prises the steps of
- providing a representative individual of the foodstuff;
determining the characteristic size of the individual,
for example length, area, volume and/or weight;
positioning the measuring lens of a colour measuring in-
strument at a surface of the foodstuff representative of the
quality parameters, and measuring the L*, a* and b* values,
and the Chroma and Hue values, of the surface;
comparing the measured values for the size of the pre-
sent individual and colorimetric data to a multivariate model
provided in advance, representative of a population of the
species of the individual, the multivariate model being
formed by mathematical processing of chemical analysis re-
sults, visual colour evaluation and individual size for the
population and comprising correlation factors between the
measured colorimetric data and a range of measured quality
parameters, in order thereby to
derive the quality parameters of the present individual
in standardized units of measurement.
The colour-measuring instrument is preferably taken from a
group consisting of colorimeter and digital camera.
CA 02642482 2010-12-21
12
The size of the individual is preferably described by means
of two or more characteristic sizes, including weight.
The chemical analysis results advantageously include the con-
tents of carotenoids and also fat and water content.
The carotenoids are preferably taken from the group consist-
ing of astaxanthin, canthaxanthin, lutein and zeaxanthin.
The visual colour evaluation is preferably indicated in a
standardized value, for example a colour card value.
Alternatively, the method includes the step of recording the
date of measuring for the individual, the multivariate model
being correlated also with the sampling dates in the popula-
tion for the chemical analysis results and the visual colour
evaluation.
Advantageously, the individual is a salmon.
Advantageously, the derived colour is indicated as a Roche
colour card value.
The invention further relates to the use of a colour-
measuring instrument for the calculation of one or more qual-
ity parameters of the foodstuff, the Chroma and Hue values
and the L*, a* and b* values being processed in a mathemati-
cal multivariate model.
In what follows is described a non-limiting example of a pre-
ferred embodiment which is visualized in the accompanying
drawings, in which:
Figure 1 shows observed seasonal variations in fat and pig-
ment for Atlantic salmon of 2-4 kg;
CA 02642482 2010-12-21
13
Figure 2 shows predicted Roche colour card values versus
read Roche colour card values for a validation set
in example 1;
Figure 3 shows predicted Roche colour card values versus
read Roche colour card values for the validation
set in example 2;
Figure 4 shows predicted astaxanthin values versus analyti-
cal astaxanthin values for the validation set in
example 2;
Figure 5 shows predicted fat values versus analytical fat
values for the validation set in example 2; and
Figure 6 shows predicted water content values versus ana-
lytical water content values for the validation set
in example 2.
For several years the applicant has analysed a large number
of salmon for chemical contents of astaxanthin, canthaxanthin
and fat, colour card values, and recorded length and weight
of the fish and date of killing of the fish. This extensive
data material has been analysed and a connection has been
found between length, weight and fat content as shown in fig-
ure 1 and quoted in the table below.
Average analysis results for Norway 1999. Salmon (Salmo
salar).
Weight class K factor* SalmoFan Astaxanthin Fat (%)
(kg) (mg/kg)
0 - 1 1.1 24.6 3.9 8.2
1 - 2 1.2 26.0 4.8 11.5
2 - 3 1.3 27.1 5.8 13.8
CA 02642482 2010-12-21
14
3 - 4 1.3 27.8 6.4 15.1
4 - 5 1.3 27.9 6.3 15.9
- 6 1.4 28.1 6.5 16.2
* K factor = (weight [g] x 100)/(length [CM]) 3
It has now surprisingly turned out that by combining a col-
orimetric measurement of the fish flesh with information
about the length and weight of the fish, the chemical con-
tents of pigments like astaxanthin and canthaxanthin, fat and
also colour card value can be predicted by means of a mathe-
matical model that takes as a starting point the L*, a* and
b* values of the colorimeter and also the length and weight
of the fish. The prediction provides a better and more reli-
able result for the chemical contents of pigment and colour
card value than the use of the values for L*, a* and b*, and
also the derived Chroma and Hue values alone, because the fat
content is taken into consideration in the prediction. At the
same time the fat content is predicted more accurately than
what may be predicted from information about the length and
weight of the fish. The basis for the prediction is a multi-
variate, statistical method leading to a set of mathematical
calibration equations.
For the accuracy of the prediction it has further turned out
to be advantageous to take into account the date of the sam-
pling as there is a seasonal variation in the colour card
measurements.
A method as described also makes it possible to use a port-
able colour-measuring instrument like a digital camera or a
colorimeter in accordance with the object of the invention.
Such instruments and the colorimeter in particular are not
dependent on standardized light conditions and make use of
built-in means for recalibration and are therefore independ-
CA 02642482 2010-12-21
ent of external means of calibration. For the calculation of
the chemical contents of pigment, colour card value and
chemical contents of fat the calibration equations are used.
These will be part of the software of a computer, for example
a laptop which may be located adjacent to the colorimeter, or
of a central computer which is reached via, for example, an
Internet interface. The read L*, a* and b* values are used as
input values together with the measured length and weight of
the fish and date of sampling. It will thereby be possible to
carry out the desired prediction immediately. The results may
be recorded manually in a separate form for example, or elec-
tronically.
The results will be provided there and then, without any fish
having to be sent to a centrally located analysing instrument
or laboratory. This will enable, for example, analysis,
evaluation and subsequently the choice of the optimum fish
feed when the farmer and feed consultant meet at the farming
facility. A better and quicker choice can then be made. Simi-
larly, the invention will contribute to an electronic colour
card measurement which may serve as objective documentation
of the fish in buying and selling.
It is obvious that the present invention could be used when
determining certain quality parameters also of other food-
stuffs. Thus, the invention is not limited to comprising
salmon only.
Instruments and software referred to in what follows, are
used in accordance with a practice normal to a skilled per-
son.
By a method according to the invention multivariate calibra-
tion techniques are applied to a combination of easily acces-
sible data from a colorimeter instrument and information
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
16
on physical data of a measured individual, for salmon
characteristic quantity data like length and weight, and
sampling date used in the models to incorporate the relevant
seasonal variations for the parameters calculated.
The surprising effect is that in the method according to the
invention a digital camera or a colorimeter can be used for
purposes, for which it is basically not well suited.
The examples that follow, exclusively describe experiments
carried out on stocks of reared salmon (Salmo salar).
Normally, fat and partially pigment content (amount of colour
in the fish) will not be measured well in visible light,
which they can be in near-infrared light (NIR). But in
combination with physical parameters'this becomes much
better. Also the determination of colour card values will be
somewhat better when data are combined as suggested.
Example 1
In the example was used a Minolta Chroma meter CR-300 type
colorimeter with a lens diameter of 0.8 cm.'Astaxanthin was
determined chemically by an HPLC method, fat was determined
chemically by Soxhlet, and water content was determined by
storing in a hot cabinet at 103 C for 16 hours. Colour was
determined visually by placing the sample in a Skretting
illumination box, that is the previously mentioned Salmon
Colour Box (Skretting AS, Stavanger, Norway) and then
comparing the colour to Roche colour card, scale from 20 to
34. The colour-level in fish varies in relation to where in
the fish it is measured. A colour card measurement and
measurement with colorimeter were carried out in a
standardized area lying in the so-called Norwegian Quality
Cut. Per definition, this is produced by cutting the fish
right behind the dorsal fin ("chop cut"). The area between
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
17
the vertebra and dorsal fin was measured. Alternatively,
measuring can be done on a fillet at a corresponding place in
the fillet, that is to say right behind the dorsal fin and
above the vertebra. Weight (round fish) and length were
recorded for each fish.
Altogether, data from 753 fish were included in the example..
145 randomly picked samples thereof were included as an
external validation set. As data in different units of
measurement (for example grams and centimetres) are included,
the data were standardized by dividing each datum by its
associated standard deviation. As a regression model was used
Partial Least Squares regression (PLS). An estimate of the,
error of prediction is given as the RMSEP (Root Mean Square
Error of Prediction):
n ~~))
/(Yi Ji)
RMSEP = ii i=1
3Z
in which "i" is the sample, "n" is the number of samples in
the set of data, "yi" is the analytical value of the sample
"i" and "yi" is the predicted value of the sample "i".
After the first modelling step approximately 5 % of the
values were considered to be "outliers", that is to say
values that were abnormal relative to the great majority of
samples in the data set, and were removed from the data set
before further modelling. The end model is based on four
significant principal components using 92 % of the variance
in the data set (L*, a*, b*, weight, length) to explain 83 %
of the variance in colour card readings. A further analysis
shows that the most important, positively correlated factors
were the coefficients of a* and b*. Weight correlated
negatively, whereas L* and length both contributed by a
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
18
smaller negative coefficient to the model. Figure 2 shows the
result of the prediction versus the measured Roche colour
card values for the validation set. The error of prediction,
expressed as the RMSEP, was 1.3 units. This is somewhat high,
but still satisfactory on the basis that the reference value
is measured subjectively by means of Roche colour cards and
that the lens of the colorimeter is somewhat small, which
limits the area measured.
Example 2
In the example 118 composite samples are included. The value
of weight, length and Roche colour card value is an average
value for each of the samples of the composite, whereas
astaxanthin, fat and water content were determined on the
composite sample as such.
= Prediction of Roche colour card value
All values were standardized. Three of the samples were
considered to be'outliers and were removed from the material.
The end model is built on three principal components
utilizing 98 % of the variance in the data for weight,
length, L*, a* and b* to explain 95 % of the variance in the
colour card values. The most important parameters of the
model correlating positively, are a* and b*. L* correlated
negatively, and length and weight correlated negatively by a
small coefficient of regression.
Figure 3 shows predicted value versus read Roche colour card
value for the validation set. The error of prediction
expressed as the RMSEP was 0.7 units, which is very good, in
particular when seen in relation to the fact that the
reference value results from a subjective reading.
= Prediction of astaxanthin content
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
19
All values were standardized. Three of the samples were
considered to be outliers and were removed from the material.
The end model is built on two principal components using 90 %
of the variance in the data for weight, length, L*, a* and b*
to explain 94 % of the variance in chemical astaxanthin
content. The most important parameters of the model
correlating positively, are a* and b*. L* correlated
negatively, whereas length and weight correlated positively,
but by a somewhat smaller contribution than a* and b*.
Figure4 shows the predicted value versus analytical
astaxanthin value for the validation set. The error of
prediction expressed as the RMSEP was 0.6 mg/kg, which is
very good.
= Prediction of fat content
All values were standardized. Seven of the samples were
considered to be outliers and were removed from the material.
The end model is built on four principal components using 98
% of the variance in the data for weight, length, L*, a* and
b* to explain 93 % of the variance in chemical fat content.
The most important parameter of the model correlating
positively is weight. Length has a negative. coefficient of
regression. The coefficients of regression of L* and a* is
smaller and negative whereas that of b* is smaller and
positive.
Figure 5 shows the predicted value versus analytical fat
value for the validation set. The error of prediction
expressed as the RMSEP was 0.8 %, which is very good.
= Prediction of water content
All values were standardized. Four of the samples were
considered to be outliers and were removed from the material.
CA 02642482 2008-08-07
WO 2007/091895 PCT/N02007/000034
The end model is built on four principal components using 98
% of the variance in the data for weight, length, L*, a* and
b* to explain 90 % of the variance in water content. The most
important parameter of the model correlating negatively is
weight. Length has a positive coefficient of regression. The
coefficients of regression of L* and b* is smaller and
positive whereas that of a* is smaller and negative.
Figure 6 shows the predicted value versus analytical fat
value for the validation,set. The error of prediction
expressed as the RMSEP was 0.8 %, which is very good.
Even though,, in the above, there are described in the main
methods of determining the quality criteria colour, pigment,
fat and water content in salmon, it is obvious that the
method could be used on other species of fish and other
species of animals, in which such quality parameters are
descriptive of product quality. It is also obvious that the
method could be used for quality assessment of, for example,
fruit. It also lies in the nature of the case that the method
could be used in calculating other quality parameters than
those mentioned here. Multivariate models for different needs
are worked out according to the same methodology as that
described above, and colorimeter measurement data and
characteristic individual-data are inserted into the
multivariate model, so that desired quality parameter
quantities are generated.
