Application of NIR technology in the animal food industry

Abstract

The importance of NIR technology in the animal food industry is presented in this study. As the example of the calibration procedure of NIR devices a calibration model for 14 samples of soybean cake was designed. Samples were previously analyzed in the standard laboratory testing of the moisture content, content of crude proteins, crude fats and crude fibre. In this calibration procedure high determination coefficients -R 2 were established for these parameters of the nutritional value of food (0.9783 for moisture, 0.9904 for crude proteins, 0.9872 for crude fats and 0.9351 for crude fibre). The comparison of values obtained by using standard laboratory methods with values obtained by NIR technology/method indicates that by using NIR devices it is possible to obtain highly reliable results, and therefore it can be used successfully in facilities for production of animal food in the control of the quality and projection of mixtures.

Full text

Biotechnology in Animal Husbandry 27 (4), p 1811-1817 , 2011 ISSN 1450-9156
Publisher: Institute for Animal Husbandry, Belgrade-Zemun UDC 636.085
DOI: 10.2298/BAH1104811M
APPLICATION OF NIR TECHNOLOGY IN THE
ANIMAL FOOD INDUSTRY
M. Maslovarić
1
, R. Jovanović
1
, S. Janković
1
, J. Lević
2
, N. Tolimir
1
1
Institute of Science Application in Agriculture, 11 000 Belgrade, Republic of Serbia
2
Institute of Food Technologies – FINS, 21000 Novi Sad, Republic of Serbia
Corresponding author: mmaslovaric@ipn.bg.ac.rs
Original scientific paper
Abstract: The importance of NIR technology in the animal food industry
is presented in this study. As the example of the calibration procedure of NIR
devices a calibration model for 14 samples of soybean cake was designed. Samples
were previously analyzed in the standard laboratory testing of the moisture content,
content of crude proteins, crude fats and crude fibre. In this calibration procedure
high determination coefficients - R
2
were established for these parameters of the
nutritional value of food (0.9783 for moisture, 0.9904 for crude proteins, 0.9872
for crude fats and 0.9351 for crude fibre). The comparison of values obtained by
using standard laboratory methods with values obtained by NIR
technology/method indicates that by using NIR devices it is possible to obtain
highly reliable results, and therefore it can be used successfully in facilities for
production of animal food in the control of the quality and projection of mixtures.
Key words: NIR, animal food
Introduction
Adequate nutrition of animals includes knowledge of the characteristics of
used feeds (Jovanovic et al., 2009). Also, intensive rearing of farm animals and
stringent requirements in regard to the quality and quantity of animal products,
require constant monitoring of the chemical composition of food for animals. On
the other hand, strong competition in the field of production of animal food,
imposes higher efficiency of the production process. Results obtained in the
standard chemical analysis methods of major quality parameters of raw materials
and finished mixtures are often received too late to stop the production process on
time or correct it (Vries et al., 2010). Therefore, it is necessary to use fast analytical
methods which enable fast response in cases when certain deviations are observed
from the projected composition or quality of the product. In this way it is possible
to achieve significant savings in the production process and ensure product of
stable quality. NIR method of analysis is exceptionally fast (usual duration of

M. Maslovarić et al.
1812
analysis is 10 seconds), non-destructive and does not require preparation of the
sample for analysis.
Main principle of NIR method of analysis. NIR(S) method (Near Infrared
Reflectance Spectroscopy) is based on screening of sample by using near infrared
reflectance/light, resulting in the spectrum of this individual sample. Spectrum can
represent dependence of the reflected or transmitted radiation from the wave
length. Results of the indirect NIR method, i.e. spectrum data, subsequently have to
be transformed into required results – concentration of the relevant constituent or
functional property of the tested material, which is obtained by
developing/designing of the calibration model. Calibration of NIR devices is done
by using samples of known composition and properties determined by standard
(reference) analysis methods. Calibration procedure includes application of certain
mathematical and statistical techniques (chemometrics) for the purpose of
obtaining of empirical equation which connects spectrum data with data obtained
by chemical analysis (Stuth et al., 2003). The costs associated with chemical
analysis of large number of samples (minimum 50, whereas approx. 150 chemical
analyses of different samples is required for open calibration sample population)
are typical for single chemometric method. Calibration of NIR devices requires
certain amount of time, but it is facilitated by the possibility of sophisticated
software packages which provide for the user application of chemometric
calibration techniques (MLR Multiple Linear Regression, PLS – Method of Partial
Least Squares Regression, PCR – Principal Component Regression, ANN –
Artificial Neural Network, etc). However, there are multiple benefits/advantages
provided by the NIR method which are beyond the time and money invested in its
implementation.
Development of calibration model represent key step in successful
implementation of NIR method and it comprises following phases:
- Selection of calibration set of samples; robustness and accuracy of the
calibration model greatly depend on the variability of the calibration population in
sense of presence of samples of different varieties, samples of various maturity
stages, samples originating from different cultivating regions and different
production years (Tsuchikawa, 2007);
- Collection of spectrum and reference data;
- Execution of the regression (calibration) model and,
Validation of the model; objective of the calibration model validation is to assess
its predicting abilities in a routine application (Petersen, 2007; Boysworth and
Booksh, 2008). We distinguish two types of validation procedure of NIRS
calibration model; cross-validation and external validation, i.e. validation using an
independent set of samples.
Both procedures result in prediction error, i.e. error which can be expected
in a routine application of model used to determine the quality, i.e. efficiency of the
developed model (Esbensen, 2006), and which is based on the concept of

Application of NIR technology in the ...
1813
differences between NIR results and results of reference laboratory analyses
(Isaksson and Segtnan, 2007; Shenk et al., 2008).
Materials and Methods
Samples of soybean cake were analyzed using standard chemical methods
(reference methods) and NIR method, in order to compare results of these two
methods. Analyses were done in FSH Komponenta from Ćuprija. The analysis of
the moisture content, content of crude fats and proteins was done on 14 samples,
and analysis of crude fibre content on 9 samples of mentioned feed. All analyses
were done according to the Rulebook on methods of physical, chemical and
microbiological analysis of animal food from year 1987. Analysis of the moisture
content was done on the laboratory moisture - meter OHAUS
®
type MB45 for fast
determination of moisture content. Chemical analysis of crude proteins was done
according to method by Kjeldahl; for determination of the crude fats content the
procedure of dry extraction according to Soxhlet method was applied, using
VELP® Scientifica digester and fat extraction apparatus. In chemical analysis of
crude fibre the Weender method was used. Chemical analysis was done in two
repetitions and average value of obtained results was considered in the study. For
NIR analysis of samples NIR device was used - PERTEN Diode Array 7200, with
rotating dish for measuring of samples, of diameter of 75 mm. This type of NIR
device does not require previous grinding of samples, which is also an advantage.
The wave length range from 950 to 1650 nm was used for measuring. The
following mathematical transformations were carried out, on the spectrum of the
soybean cake sample, in order to remove all irrelevant information from the
spectrum, and to be able to interpret the results easier: Savitzky-Golay – to obtain
first spectrum excerpt and MSC – Multiplicative scatter correction, and for the
purpose of development of calibration model, PLS1 – Partial Least Squares
Regression. For transformation of the main spectrum of the soybean cake samples
and execution/development of calibration model software package CAMO
Unscrambler
®
10.1 was used.
Results and Discussion
In assessment of the efficiency of the obtained calibration model the
following statistical tests were used: RMSEP – Root Mean Square Error of
Prediction and R
2
– Determination coefficient R
2
(Table 1). Root Mean Square
Error of Prediction – RMSEP represent measure of the variability of differences
between values obtained by NIR analysis (predicted values) and reference
laboratory methods. When the value of RMSEP is closer to the zero, the calibration
model is more reliable. Determination coefficient R
2
, represents the square of the
correlation coefficient – r, and describes the variation and range of the calibration

M. Maslovarić et al.
1814
set. Value of R
2
= 1, means that 100% of variations are described with the
calibration (Pojić, 2010). Based on presented results it can be concluded that very
good determination coefficients (R
2
) in all cases are obtained. In comparison of
concentration values of tested parameters obtained by standard laboratory method
and those obtained by NIR method (Table 2), it is observed that results of the NIR
method of analysis show no significant deviations from results obtained by
standard laboratory methods, and errors which occurred did not exceed deviation
regulated in the Rulebook on quality and other requirements for animal food
(„Official Journal of RS“, no. 4/2010). The highest determination coefficient (R
2
)
was achieved in crude proteins, which was expected since N-H chemical bond
present in proteins shows high level of absorption of infrared radiation (Oatway et
al, 2006). Very good determination coefficient (R
2
) was also achieved in crude fats,
although the concentration range of this constituent was considerably wide.
Slightly lower determination coefficients (R
2
) were obtained fro moisture and
crude fibre. On the other hand, value of the Root Mean Square Error of Prediction -
RMSEP was the lowest in crude fibre (lower than in crude proteins) which is
explained by very narrow range of content of this parameter.
Table 1. Concentration of crude proteins and crude fats in the calibration set (%), range of
their values and results of the statistical cross validation
Average of
values, %
SD Range R
2
RMSEP
Moisture 7.31 1.52 4.06 – 9.65 0.9783 0.2442
Crude proteins 38.12 0.53 37.29 - 39.15 0.9905 0.0517
Crude fats 7.95 1.21 6.49 - 11.31 0.9884 0.0972
Crude fibres 6.02 0.32 5.72 – 6.91 0.9351 0.04331
SD – Standard deviation, R
2
– Determination coefficient, RMSEP – Root Mean Square Error of
Prediction.
Table 2. Comparative review of the results of standard laboratory and NIR methods of analysis
Moisture Crude proteins Crude fats Crude fibres
Sample
St.l.m, % NIR, % St.l.m, % NIR, % St.l.m, % NIR, % St.l.m, % NIR, %
1 4.060 4.446 37.780 38.669 8.880 8.761 5.720 5.742
2 7.540 7.752 38.140 38.202 7.000 7.131 6.080 6.001
3 7.000 6.849 37.720 37.674 8.880 8.985 5.900 5.927
4 8.080 8.373 37.810 37.816 7.180 7.082 5.880 5.882
5 8.160 8.047 37.400 37.391 7.290 7.406 5.920 5.952
6 9.130 8.921 37.330 37.349 6.490 6.533 5.680 5.705
7 7.130 7.143 37.800 37.798 8.090 8.061 6.020 6.002
8 8.160 8.266 37.770 37.791 7.950 7.858 6.100 5.952
St.l.m – Value obtained by standard laboratory method
NIR – Value obtained by NIR method of analysis

Application of NIR technology in the ...
1815
During the process of development of calibration model and its validation,
certain number of samples had to be excluded in order to obtain the most reliable
model possible. Therefore, the number of samples presented in Table 2 is lower
than initial number of samples of soybean cake (9 samples for crude fibre and 14
samples for other parameters). This is also reason why, in case of moisture, crude
proteins and crude fibre, a model was obtained which gives more reliable results,
but with narrower range of contents of determined nutritional parameters. In case
of crude fats, however, a calibration model was obtained which gives very reliable
results for wide range of concentrations of this constituent. Based on obtained
results it can be concluded that even with small number of samples which were
tested by using standard laboratory methods of analysis, it is possible, using NIR
method, to receive very reliable results, but adequate selection of samples for
development of calibration model is of crucial importance. When the number of
samples with various concentrations of relevant constituents is higher, a calibration
model for wider range of concentrations can be developed, which would also give
reliable results.
There are also other ways of implementation of NIR technology, such as
use of calibrations offered/provided by manufacturers of NIR devices. However,
this way is associated with considerable costs. Also, calibration curves provided by
manufacturers of NIR devices, although developed on large number of samples,
often need to be adjusted due to high variations in the quality of raw materials.
Conclusion
Introduction of NIR technology is a good strategy of development of
facilities for production of animal food, if the complexity of their production
process is taken into consideration. NIRS method, as fast analytical method,
enables timely receiving of required results, improvement in the quality control and
increase in the efficiency of the animal food production, wit final outcome in
increase of the profitability and competitiveness.
Results obtained by applying NIR method in analysis of the soybean cake
samples showed no significant differences compared to results obtained by
standard laboratory methods, which proves that NIR method can be very reliable in
determination of the composition of raw materials used in production of animal
food, as well as of finished mixtures. Also, duration of NIR method of analysis was
incomparably shorter than duration of standard laboratory methods.

M. Maslovarić et al.
1816
Primena NIR tehnologije u industriji hrane za životinje
M. Maslovarić, R. Jovanović, S. Janković, J. Lević, N. Tolimir
Rezime
Sve oštrija konkurencija u domenu proizvodnje hrane za životinjezahteva
da proces proizvodnje bude što brži i efikasniji. NIRS metoda, kao izuzetno brza
analitička metoda, predstavlja dobru soluciju koja poseduje veliki potencijal za
unapređenje nadzora i kontrole proizvodnog procesa u fabrikama hrane za
životinje. Uzorci sojine pogače analizirani su standardnim hemijskim metodama
(referentne metode) i NIR metodom, u cilju poređenja rezultata ove dve metode.
Analiziran je sadržaj vlage, sirovih proteina, sirove masti i sirove celuloze. U
postupku kalibracije NIR uređaja za ove parametre kvaliteta sojine pogače,
ostvareni su veoma dobri koeficijenti determinacije (R
2
), što znači da su dobijeni
pouzdani kalibracioni modeli. Rezultati dobijeni primenom NIR metode za analizu
uzoraka sojine pogače (u postupku validacije modela), nisu se bitnije razlikovali od
rezultata standardnih laboratorijskih metoda, što dokazuje da da NIR metoda može
biti vrlo pouzdana u određivanju sastava, kako sirovina koje se koriste u
proizvodnji hrane za životinje, tako i gotovih smeša. Prednosti uvođenja NIR
tehnologije u fabrike hrane za životinje su: brza kontrola sadržaja relevantnih
sastojaka u sirovinama i gotovim proizvodima, poboljšanje kvaliteta proizvoda,
snižavanje troškova, mogućnost ugradnje NIR uređaja u proizvodni proces (at line
analiza) - kontinualna kontrola procesa proizvodnje hrane za životinje.
References
BOYSWORTH M.K., BOOKSH K.S. (2008): Aspects of multivariate calibration
applied to near - infrared spectroscopy. D.A. Burns & E.W. Ciurczak Handbook of
Near – Infrared Analysis, Boca Raton: CRC Press Taylor & Francis Group, 207-229.
ESBENSEN K.H. (2006): Multivariate Data Analysis In Practice. Oslo: Camo
Software AS
GIVENS D.I., DEAVILLE E.R. (1999): The current and future role of near
infrared reflectance spectroscopy in animal nutrition: a review. Aust. J. Agric. Res.,
50, 1131-1145.
DE VRIES P.F., JOVANOVIĆ R., POJIĆ M., MASLOVARIĆ
.
M.,
MILENKOVIĆ I. (2010): Značaj primene NIRS metode u fabrikama hrane za
životinje u cilju poboljšanja kvaliteta i konkurentnosti. Zbornik radova 14.
Međunarodnog simpozijuma tehnologije hrane za životinje -“Tehnologija hrane za
životinje, kvalitet i bezbednost”.

Application of NIR technology in the ...
1817
HRUSCHA W.R. (1987): Data analysis: Wavelength selection methods. In:
WILLIAMS P.C., NORRIS K.H. (eds), Near Infrared Technology in the
Agriculture and Food Industries. Am. Assoc. of Cereal Chem., St. Paul, MN, 35-55.
ISAKSSON T., SEGTNAN V.H. (2007): Meat and fish products. Y. Ozaki, W.F.
McCLURE and A.A. Christy, Near – infrared spectroscopy in Food Science and
Technology, Hoboken: John Wiley and Sons, 247-277.
JOVANOVIĆ R., GRUBIĆ G., LEVIĆ J., JANKOVIĆ S. (2009): The effect of
digestibility of maize hybrids on production performance of fattening young cattle.
Biotechnology in Animal Husbandry, 25, 5-6, 677.
OATWAY L.A., HELM J.H., JUSKIW P.E. (2006): Paper: Development of Near
Infrared Spectroscopy to Screen for Feed Quality Characteristics in Whole Grain
Barley, Field Crop Development Centre, Agriculture and Rural Development,
Government of Alberta, USA.
PETERSEN P.E. (2007): All you ever wanted to know about chemometrics – but
didn’t like to ask. In Focus, 31, 22-23.
POJIĆ M. (2010): Definisanje postupka za razvoj kalibracionog modela za blisku
infracrvenu spektroskopiju. Doktorska disertacija, Tehnološki fakultet, Novi Sad.
STUTH J., JAMA A., TOLLESON D. (2003): Direct and indirect means of
predicting forage quality trough near infrared reflectance spectroscopy. Field Crops
Res., 84, 45-46.
SHENK J.S., (2004): The wonderful world of visible-near infrared spectra: theory
and practice. A.M.C. Davies & Garrido-Varo, Proceedings of the 11
th
International
Conference on Near Infrared Spectroscopy, Chichester: NIR publications, 33-40
TSUCHIKAWA S. (2007): Determination of dry matter content and basic density
of Norway spruce by near infrared reflectance and transmittance spectroscopy.
Journal of Near Infrared Spectroscopy, 2, 127-135
VRIES P. F., JOVANOVIĆ R., POJIĆ M., MASLOVARIĆ
.
M., MILENKOVIĆ I.
(2010): Značaj primene NIRS metode u fabrikama hrane za životinje u cilju
poboljšanja kvaliteta i konkurentnosti, Zbornik radova 14. Međunarodnog
simpozijuma tehnologije hrane za životinje -“Tehnologija hrane za životinje,
kvalitet i bezbednost”, 304, 315.
Received 30 June 2011; accepted for publication 15 August 2011