Content uploaded by M. S. Islam
Author content
All content in this area was uploaded by M. S. Islam on Dec 18, 2018
Content may be subject to copyright.
316
International Journal of
Agriculture and Biosciences
www.ijagbio.com P-ISSN: 2305-6622 E-ISSN: 2306-3599 editor@ijagbio.com
RESEARCH ARTICLE
Nutritional Evaluation of Commercial Broiler Feeds by Using Near Infrared Reflectance
Spectroscopy
*1Khaleduzzaman ABM, ME Haque2 and MS Islam3
1Animal Nutrition Laboratory, Department of Livestock Services (DLS), Ministry of Fisheries and Livestock (MoFL),
Farmgate, Dahaka 1215, Bangladesh; 2Department of Animal Nutrition and Livestock Management, Faculty of
Veterinary and Animal Science, Sylhet Agricultural University, Sylhet, Bangladesh; 3Agro Technology Discipline,
Khulna University, Khulna 9208, Bangladesh
ARTICLE INFO
ABSTRACT
Received:
Revised:
Accepted:
August 12, 2013
September 23, 2013
October 10, 2013
Broiler feeds from different commercial feed mills were collected and NIR
spectrum of feed samples were obtained in duplicate (scanning number 32,
resolution 8) with an FT-NIRS (Bruker, MPA, Germany) systems mono-
chromator (700-2400 nm) using a Quartz cup sampling device. Multivariate
analysis were performed for the development of calibration equations of
nutrient content by an Optical User Software (OPUS) and Opus-Lab to relate
the spectral data and corresponding concentration values of broiler and layer
feeds. Data were centered using Partial Least Squares (PLS) algorithm and
spectral outliers were identified from each calibration. The calibration models
were validated by RMSECV (Root Mean Square Error Cross Validation),
RMSEE (Root Mean Square Error of Estimation) and correlation coefficient (r2)
between the measured values of nutrient component determined by analytical
laboratory versus predicted values by the NIRS. The standard error of
estimation (RMSEE) for the determination of moisture, CP, CF, EE, Ca and P
in broiler feeds was 0.230, 0.351, 0.361, 0.350, 0.056 and 0.021% respectively
with correlation coefficient (r2) of 86.09, 95.77, 86.28, 96.28, 80.50 and 95.80.
After cross validation, the standard error (RMSECV) for the prediction of
moisture, CP, CF, EE, Ca and P in broiler feeds was 0.242, 0.371, 0.406, 0.390,
0.066 and 0.031% respectively having correlation coefficients (r2) of 84.32,
92.20, 81.84, 95.46, 75.00 and 94.71. From the results of the present study it
may be concluded that NIRS could potentially be used to predict the moisture,
CP, CF, EE, Ca and P contents in commercial broiler feeds.
Key words:
Broiler feeds
NIRS
Nutrient evaluation
Quality
*Corresponding Address:
ABM Khaleduzzaman
abmk5566@gmail.com
Cite This Article as: Khaleduzzaman ABM, ME Haque and MS Islam, 2013. Nutritional evaluation of commercial
broiler feeds by using Near Infrared Reflectance Spectroscopy. Inter J Agri Biosci, 2(6): 316-320. www.ijagbio.com
INTRODUCTION
Broiler farming is one of the fastest growing and
most promising industries in Bangladesh. Its steady
growth (15-20%) results in attaining country’s economic
growth which also contributes to improve the nutritional
status by supplying meat. The government is getting
interested in broiler farming and is encouraging both
urban and rural people to work here and enhance capacity.
Intensive production system, however, depends solely on
compound feeds the cost of which represents 65-70% of
the total cost of broiler production. However, proper
attention should be given to evaluate the nutritional
quality of feed ingredients in order to supply the adequate
amount of balanced diet to poultry for maximizing the
productivity cost effectively.
Near Infrared Reflectance Spectroscopy (NIRS)
offers the potential for obtaining a rapid, nondestructive
and accurate estimate of the chemical composition of
feedstuffs. The technique has extensive application for the
analysis of constituents of agricultural crops, feeds and
foods (Osborne et al., 1993; Williams and Norris, 2001;
Roberts et al., 2004). NIR spectroscopy, while much
simpler and more rapid than traditional analytical methods
typically requires grinding a sample to a fine particle size
to give a smooth and homogeneous surface for reflection
and increased precession. This analytical tools requires no
chemical reagents, therefore, avoids the problems of
organic and other chemical waste disposal. Once
calibrations are in place, it takes just minutes to have the
result of one or more constituents which by conventional
chemistry may take hours or days. Currently, NIRS of
Inter J Agri Biosci, 2013, 2(6): 316-320.
317
whole grains at the grain elevator is used widely in the
USA, Canada, Australia and Europe for evaluation of
protein and moisture content of grains (Sandra et al.,
2005). However, little is known about the potential of NIR
spectroscopy for the nutritional evaluation of locally
available ingredients in Bangladesh as well as other parts
of Southeast Asia and quick prediction of nutritional
quality of feed ingredients in the feeds is necessary for
achieving sustainable poultry production. Therefore, the
aim of this study was: a) to develop calibration procedures
using NIRS and validation of calibrations for the evaluation
of locally available broiler feed samples accurately and b)
to determine the nutritive value of large number of
commercial broiler feeds within shortest possible time.
MATERIALS AND METHODS
Sample collection and preparation
Broiler feed samples were collected from the
different commercial feed mills and feed markets
available in Bangladesh from 2006 to 2009. Samples were
ground through 2.0 mm screen for the homogeneous
particle size using Cemotec Grinding Mill (Foss Tecator,
Sweden) before reflectance measurements and dried at 600
C for wet chemistry analysis.
Spectroscopic and laboratory analysis of the sample
Identification of appropriate samples was the first
step in utilizing a combination of NIRS and laboratory
analysis. Ground broiler feed samples were scanned in
duplicate (scanning number 32, resolution 8) with an FT-
NIRS (Bruker, MPA, Germany) systems monochromator
(700-2400 nm). Samples were packed into a Quartz cup
sample holder which holds the sample into a clear glass
window and to maintain good contact between the
granular sample and window. Monochromatic light was
focused on to a concave mirror which separated the light
into its composing wavelengths (Figure 1). Required
wavelength was selected by which the near infrared light
was fallen on to the feedstuff. The amount of this light
which was reflected by the feedstuff was measured to
obtain the absorption corresponding to selected wave-
length. The graph depicting the wavelength tested versus
the absorption is called the spectrum, and an example
spectrum is presented in the top right corner (Figure 1).
Spectral reference curves were measured each of the
broiler feed sample and the data were stored in selected
folder. Sample moisture, CP, CF, EE, Ca and P were
determined according to the procedure of AOAC (2000).
Development of calibration model
For the development of calibration model in the
present experiment, multivariate analysis was performed
by a commercial analysis program Optical User Software
(OPUS) and Opus Lab provided by Bruker, MPA,
Germany to relate the spectral data and corresponding
concentration values for each nutrient component
(Moisture, CP, CF, EE, Ca and P) of broiler feed samples.
The model was developed using Partial Least Squares
(PLS) algorithm and the spectral data were processed by a
suitable mathematical method e.g. first derivative, second
derivative, vector normalization, subtraction of straight
line etc. PLS uses eigenvectors and eigenvalues to
Fig. 1: Schematical layout of NIRS machine. Light is focused
(1) on to a concave mirror (2), which separates the light into its
composing wavelengths (3). One wave-length is selected (4),
and falls onto the feedstuff (5). The amount of this light which is
reflected by the feedstuff is measured (6) to obtain the
absorption corresponding to one wavelength. By changing the
orientation of the mirror (2), different wavelengths of light are
selected to obtain absorption measures for all wavelengths of
interest. The graph depicting the wavelength tested versus the
absorption is called the spectrum, and an example spectrum for
coconut meal is presented in the top right corner (7).
perform a decomposition of the spectral and constituent
concentration data simultaneously. The decomposition
process is a systematic means to determine the most
important variations in the data. PLS uses constituent
concentration information during spectral decomposition,
which weights spectra containing higher constituent
concentrations more heavily. The term “factor” or “rank”
is used to describe a linear combination of spectral data.
PLS reconstructs a spectrum that represents the predicted
constituent values. This predicted spectrum is subtracted
from the actual spectrum to determine residuals.
Therefore, the residual (Res) is the difference between the
true and fitted value. Thus the sum of squared errors (SSE)
is the quadratic summation of these values:
SSE = ∑[Resi]2
The standard error of calibration or root mean square
error of estimation (RMSEE) is calculated from this sum,
with M being the number of standards and R the PLS
factors or rank:
R
MSEE = √ 1 SSE
M-R-1
Appropriate frequency range of the spectrum was
selected to get good correlation between the changes in
spectral and the concentration data.
Validation of the calibration model
The suitability of the chosen data processing method
and the frequency range for method development was
evaluated during validation. In the case of cross
validation, individual samples were taken from the
calibration set. Using the remaining samples, a calibration
model was established and used to analyze the previously
extracted samples. This procedure of removing samples,
Inter J Agri Biosci, 2013, 2(6): 316-320.
318
Table 1: General statistics for the chemical composition of
broiler feed samples
Nutrients Sample
No Min.-Max. Mean SE
1
%
Moisture 543 10.90-13.11 12.23 0.023
CP 352 13.12-22.96 17.96 0.090
CF 215 1.78-7.20 4.05 0.065
EE 374 3.09-11.35 7.02 0.094
Ca 230 0.63-1.27 0.94 0.007
P 241 0.38-0.83 0.66 0.006
1SE = Standard error
Fig. 2: Steps in cross validation
analyzing them, and returning them to the calibration data
set was continued successively until all the samples had
been analyzed once (Figure 2).
A comparison of the resulting analysis values with
the original raw data allowed the calculation of the
predictive error of the complete data system, the root
mean square error cross validation (RMSECV):
RMSECV = √ 1 M ∑ (Differi)2
M
i=1
Besides, coefficient of determination (r2) from the
linear regression of measured values of nutrient
component determined by analytical laboratory versus
predicted values by the NIR calibration was calculated to
give the accuracy of the model. During the validation,
potential outliers could be detected easily and only after
all outliers had been removed from the calibration data
set, and finally after the optimum parameters had been
found, the calibration model was established.
RESULTS AND DISCUSSION
The overall range of moisture, CP, CF, EE, Ca and P
contents in broiler feeds were 10.90 to 13.11, 13.12 to
22.96, 1.78 to 7.20, 3.09 to 11.35, 0.63 to 1.27 and 0.38 to
0.83 with standard error (SE) of 0.023, 0.090, 0.065,
0.094, 0.007 and 0.006 respectively (Table 1). The
number of samples in developing calibration equations for
the prediction of CF (n=215), Ca (n=230) and P (n=241)
were relatively lower than the number of samples used in
developing equations for predicting moisture (n=543), CP
(n=352) and EE (n=374). The SE of CP, CF and EE
contents in broiler feeds were slightly higher than that of
moisture, Ca and P contents which probably due to the
higher variations of CP, CF and EE contents in broiler feeds.
The standard error of estimation (RMSEE) in
developing calibration equations by using NIRS for the
determination of moisture, CP, CF, EE, Ca and P of
broiler feeds were 0.230, 0.351, 0.361, 0.350, 0.056 and
0.021% respectively with correlation coefficient (r2) of
86.09, 95.77, 86.28, 96.28, 80.50 and 95.80 (Table 2).
After cross validation, the standard error (RMSECV) for
the prediction of moisture, CP, CF, EE, Ca and P in
broiler feeds were 0.242, 0.371, 0.406, 0.390, 0.066 and
0.030% respectively. Besides, the correlation coefficients
(r2) between measured values of nutrient component
determined by analytical laboratory versus predicted
values determined by the NIR calibrations for the
determination of moisture, CP, CF, EE, Ca and P in
broiler feeds were 84.32, 95.20, 81.84, 95.46, 75.00 and
94.71 respectively (Table 2 and Figure 3). The standard
error in calibration (RMSEE) and after cross validation
(RMSECV) for the prediction of CP, CF and EE contents
in broiler feeds were slightly higher than that of
calibrations for the prediction of moisture, Ca and P in
broiler feeds which could be due to higher variations
observed in laboratory determinations of CP, CF and EE
contents in boiler feeds (Table 1). Holechek et al. (1982)
reported that the high errors of analysis could be due to
differences in botanical composition, maturity of grains,
method of collection and preparation of samples as well as
host environmental interactions. However, the correlation
coefficient (r2) between measured CP, CF and EE contents
of broiler feeds in the laboratory and predicted nutrient
contents by NIR calibration equations were 95.77, 86.28
and 96.28 were satisfactory. Similarly, after cross
validation, the correlation between laboratory values and
NIR predicted values of CP, CF and EE in broiler feeds
were satisfactory also (Figure 3b, 3c and 3d). Melchinger
et al. (1986) found the SE of prediction and correlation
coefficient (r2) for the prediction of CP in maize grains to
be 0.29% and 96.00 respectively. Therefore, the model
developed for CP, and EE determination in broiler feed
samples in the present experiment appears to be
sufficiently accurate and latter for quality control
applications. Besides, the correlation coefficient ((r2) for
the prediction of CF contents in broiler feeds could be
improved by the addition of large number of samples in
calibration set.
In predicting moisture, Ca and P contents in broiler
feeds, the RMSEE were 0.023, 0.056 and 0.021%
respectively and after cross validation the RMSECV were
0.242, 0.066 and 0.031% respectively. The SE for the
prediction of moisture, Ca and P by using NIRS were
almost similar nature to the SE of laboratory
determination of moisture, Ca and P in broiler feeds.
According to the procedure of Conzen (2003),
multivariate calibration was developed and in the present
experiment the commercial software analysis program
OPUS (Optical User Software) had the opportunity to use
all possible mathematical models i. e. first derivative,
second derivative, vector normalization etc. to develop
calibration equations with least errors. In addition, the
correlation coefficient (r2) between measured moisture
and Ca contents of broiler feeds in the laboratory and
predicted moisture and Ca contents by NIRS after cross
validation were 84.32 and 94.71 respectively indicate the
accuracy of the model (Figure 3a and 3e). However, the
Inter J Agri Biosci, 2013, 2(6): 316-320.
319
Table 2: NIR statistics for the prediction of moisture, CP, CF, EE, Ca and P contents in broiler feeds
Nutrients Sample No Cross validation statistics Calibration statistics
Rank RMSECV1 r
2 Rank RMSEE2 r
2
Moisture 543 08 0.242 84.32 08 0.230 86.09
CP 352 05 0.371 95.20 05 0.351 95.77
CF 215 09 0.406 81.84 09 0.361 86.28
EE 374 08 0.390 95.46 08 0.350 96.28
Ca 230 10 0.066 75.00 10 0.056 80.50
P 241 10 0.031 94.71 10 0.021 95.80
1RMSECV = Root Mean Square Error Cross Validation; 2RMSEE = Root Mean Square Error of Estimation; r2 = correlation coefficient
(a) (b)
(c) (d)
(e) (f)
Fig. 3: Prediction (NIR) vs true (laboratory), RMSECV, PLS factors (Rank) and correlation coefficient (r) for the prediction of (a)
moisture, (b) CP, (c) CF, (d) EE, (e) Ca and (f) P determination in broiler feed samples.
Inter J Agri Biosci, 2013, 2(6): 316-320.
320
correlation coefficient (r2) of P between laboratory and
NIRS determination was 75.00 (Figure 3f) which was
relatively lower after cross validation. In PLS regression,
if too many factors (rank) are chosen (10 for Ca) the
model tries to account even the smaller changes in data set
which creates spectral noise (“overfitting”) and leads to
decrease correlation coefficient after validation (Conzen,
2003). However, this could be increased by incorporation
large number of homogeneous samples in calibration set.
Conclusions
In the present experiment, the samples were collected
from different feed mills and locations having variations
in nutrient contents and thus the improvement of
calibrations has not been as great as that observed with
homogeneous calibrations. The correlation coefficient (r2)
for the prediction of CF and Ca contents in broiler feeds
could be improved by the addition of large number of
homogeneous samples in calibration sets. However, from
the above discussion it can be said that it is obligatory to
predict animal feeds for quick manufacturing process.
NIRS should be preferred as the most convenient tool in
this regard, because this non-destructive analytical method
needs no chemical reagents and hence it is environ-
mentally sound enough as it does not have any organic or
chemical waste.
REFERENCES
AOAC, 2000. Official methods of analysis. Association of
official analytical chemist, 17th ed, Washington DC,
USA.
Conzen JP, 2003. Multivariate calibration. A practical
guide for developing methods in the quantitative
analytical chemistry. 1st English edition, translated
from the 3rd German edition. Bruker Optick, GmbH.
Holechek JL, JS Shenk, M Vavra, and D Arthun, 1982.
Prediction of forage quality using near infrared
reflectance spectroscopy on esophageal fistula
samples from cattle on mountain range. J Anim Sci,
55: 970.
Melchinger AE, GA Schmidt and HH Geiger, 1986.
Evaluation of near infrared reflectance spectroscopy
for predicting grain and stover quality traits in maize.
Plant Breeding, 97: 20-29.
Osborne BG, T Fearn and PH Hindle, 1993. Practical NIR
spectroscopy with applications in food and beverage
analysis. Longman Scientific and Technical, Harlow,
UK.
Roberts CA, JR Workman and JB Reeves, 2004. Near
infrared reflectance spectroscopy in agriculture.
Agronomy Monograph, 44. ASA, CSSA, and SSSA,
Madison, WI.
Sandra EK, N Shimizu, FE Barton II and K Ohtsubo,
2005. Near infrared transmission and reflectance
spectroscopy for the determination of dietary fiber in
barley cultivars. Crop Sci, 45: 2307
Williams PC, 2001. Implementation of near infrared
technology. In: Williams, PC and Norris, KH (ed)
Near-infrared technology in the agricultural and food
industries, 2nd edition, American Association of
Cereal Chemists, St Paul, MN, p: 145-169.
