Research Paper

Prediction and classication of sugar content of sugarcane

based on skin scanning using visible and shortwave near
infrared
Nazmi Mat Nawi
a,c,
*, Guangnan Chen
a,b
, Troy Jensen
a,b
, Saman Abdanan Mehdizadeh
d
a
Faculty of Engineering and Surveying, University of Southern Queensland, Toowoomba, QLD 4350, Australia
b
National Centre for Engineering in Agriculture (NCEA), University of Southern Queensland, Toowoomba, QLD 4350, Australia
c
Department of Biological and Agricultural Engineering, Faculty of Engineering, Universiti Putra Malaysia, 43400 Serdang,
Selangor, Malaysia
d
Department of Agricultural Machinery and Mechanization, Ramin Khuzestan University of Agriculture and Natural Resources,
Khuzestan, Iran
a r t i c l e i n f o
Article history:
Received 2 January 2013
Received in revised form
27 February 2013
Accepted 8 March 2013
Published online 19 April 2013
The potential application of a visible and shortwave near infrared (Vis/SWNIR) spectro-
scopic technique as a low cost alternative to predict sugar content based on skin scanning
was evaluated. Two hundred and ninety one internode samples representing three
different commercial sugarcane varieties were used. Each sample was scanned at four
scanning points to obtain the spectra data which was later correlated with its

Brix (soluble
solids content) values. Partial least square (PLS) model was developed and applied to both
calibration and prediction samples. Using reectance spectra data, the model had a coef-
cient of determination (R
2
) of 0.91 and root means square error of predictions (RMSEP) of
0.721

Brix. The articial neural network (ANN) was also applied to classify spectra data
into ve

Brix categories. The ANN has yielded good classication performance, ranging
from 50 to 100% accuracy with an average accuracy of 83.1%. These results demonstrated
that the Vis/SWNIR spectroscopy technique could be applied to predict sugarcane

Brix in
the eld based skin scanning method.
2013 IAgrE. Published by Elsevier Ltd. All rights reserved.
1. Introduction
Precision agriculture (PA) is a valuable management strategy
to enhance farm prots through efcient application of crop
inputs by matching them with variations of crop yield and
quality in the eld (Wendte, Skotnikov, & Thomas, 2001). The
primary requirements for PA application are yield monitoring
and mapping. To date, several studies have been carried out to
produce a yield map in sugarcane industries around the
world, including Australia (Cox, Harris, Pax, & Dick, 1996),
Brazil (Magalha es & Cerri, 2007) and the US (Price, Johnson,
Viator, Larsen, & Peters, 2011). However, an extensive review
published by Bramley (2009) regarding PA application in sug-
arcane industry revealed that current PA technologies only
monitor the yield but do not have the ability to measure
product quality. This is a serious limitation because both yield
* Corresponding author. Department of Biological and Agricultural Engineering, Faculty of Engineering, Universiti Putra Malaysia, 43400
Serdang, Selangor, Malaysia.
E-mail addresses: nazmi@eng.upm.edu.my, naj_miey@hotmail.com (N. Mat Nawi).
Available online at www.sciencedirect.com
j ournal homepage: www. el sevi er. com/ l ocat e/ i ssn/ 15375110
b i o s ys t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1
1537-5110/$ e see front matter 2013 IAgrE. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.biosystemseng.2013.03.005
and quality components vary signicantly across the eld
(Bramley, Panitz, Jensen, & Baillie, 2012; Johnson & Richard,
2005). An accurate determination of both components is
important especially for a fair payment systemto growers and
better farm management practices. Thus, it is desirable to
measure quality level along with yield data in the eld during
harvesting to fully utilise the PA approach.
In Australia, sugarcane quality is determined based on its
sugar content, known as commercial cane sugar (CCS). CCS is
derived from

Brix (soluble solids content), Pol (the sugar
content measured using the property of optical activity which
causes polarised light to be rotated) and bre content. The

Brix
and Pol are usually measured in the laboratory using juice
samples. Unfortunately, replicating this laboratory method in
the eld is very difcult due to the absence of reliable sensor
and the difculties obtaining sufcient juice samples. There-
fore, an ability to predict sugarcane quality by directly scan-
ning stalk skin using spectroscopic methods would be an
attractive alternative. Spectroscopic methods have widely
been applied to non-destructively predict internal crop quality
based on skin scanning of a number of crops, e.g. kiwifruit
(Moghimi, Aghkhani, Sazgarnia, & Sarmad, 2010), apples (Lu,
2004; Park, Abbott, Lee, Choi, & Choi, 2003) and pineapple
(Chia, Rahim, & Rahim, 2012; Guthrie & Walsh, 1997).
For eld application, however, a portable, robust and low
cost spectrometer is preferable (Montes et al., 2006). Lately,
the development of a portable visible and shortwave near
infrared (Vis/SWNIR) instrument with a wavelength range
from 350 to 1100 nmappears promising because the bands are
ascribed to the third and fourth overtones of OeH and CeH
stretching modes and the instrument is low-cost and portable
enough for in-eld measurements (Walsh, Guthrie, & Burney,
2000). The use of this spectrometer, coupled with the use of
multivariate statistical techniques could create the possibility
to intensively and precisely map crop quality over large parcel
of cropping land (Menesatti et al., 2010). Several studies have
been successfully conducted in the eld using a spectrometer
mounted on a harvester for different crops, including maize
(Montes et al., 2006), forage (Digman & Shinners, 2008) and
sugar beet (Panigrahi & Hofman, 2003).
Unfortunately, there are no reported studies regarding the
skin scanning method for quality prediction in sugarcane
industry. This method works based on the principle that when
a light beam hits the stalk, a small fraction is reected at the
surfaceas a specular reectanceandthe rest will penetrateinto
fruit tissues. In the tissues, photons are absorbed or migrate in
different directions where radiationwill be scattered backward
to the surface as diffuse reectance, while the remaining ra-
diation migrates forward into the tissues as absorbance (Qing,
Ji, & Zude, 2007). Light absorption is related to certain chemi-
cal constituents, such as sugar, acid, water, etc. (Williams &
Norris, 2001). A reectance spectrometer measures the aggre-
gate amount of light reected from a sample, from which light
absorption may be estimated and then related to certain
chemical constituents of the crop (Lu, 2004).
For mapping purposes, the crop quality data must also be
classied into suitable quality classes. The articial neural
network (ANN) is a well-known non-linear method which
could provide a robust classication model (Lee et al., 2010).
ANN is a multivariate statistical data analysis technique which
often gives higher recognition and prediction probability than
statistical classication algorithms. Inspired by biological ner-
vous system, the neural networks are composed of a number of
elements operating in parallel. By adjusting the weights (con-
nections) between elements, a neural network can perform
several functions, such as prediction, data ltration, data
conceptualization, classication, and data association (Haykin,
1994). The applications of ANN to classify crop quality have
been reported by Wu, Feng, Zhang, and He (2008) and Xing,
Symons, Shahin, and Hatcher (2010). In this study, ANN
models were applied to classify sugarcane quality based on
several

Brix values. Thus, the specic objectives of this study
were (1) to investigate the feasibility of using Vis/SWNIR to
predict the quality level based on skin scanning; (2) to classify
sugarcane quality based on ve

Brix classes using ANN.
2. Material and methods
2.1. Sample preparation
A total of 291 internodes were extracted from 22 sugarcane
stalk samples. The stalk samples were collected from the
research station of Bureau of Sugar Experimental Station
Nomenclature
A Absorbance spectra
ANN Articial neural network
ASCII American Standard Code for Information
Interchange
ASD Analytical spectral device
BSES Bureau of Sugar Experimental Station
CCS Commercial cane sugar
FOV Field-of-view
FWHM Full width at half maximum
H High
L Low
LVs Latent variables
M Medium
MSC Multiplicative scatter correction
PA Precision agriculture
PCA Principal component analysis
PCs Principal components
PLS Partial least square
R Reectance spectra
R
2
coefcient of determination
RMSEP Root mean square error of prediction
SEC Standard error of calibration
SEP Standard error of prediction
SNV Standard normal variate
SSC Soluble solids content
Vis/SWNIR Visible/shortwave near infrared
VisNIR Visible near infrared
b i os y s t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 155
(BSES), Bundaberg, Queensland, inMay 2012. The stalk samples
were of a commercial variety cut froma propagationblock used
to source materials for plant breeding trials. They were a plant
crop that was planted in September 2011 (8 months old). The
crop was grown under commercial conditions with the fertil-
isation based on soil test and the six easy steps nutrition
guidelines. The stalks belong to commercial variety trials rep-
resenting three different maturity stages, namely early-
maturing (Q155), mid-maturing (Q208) and late-maturing
(Q190) crops. The selection of these three varieties was to
ensure that the models developed in this study cover a wide
range of

Brix (7.6e22.2) which is representative during har-
vesting. Whole stalk samples were rst topped and cut into
individual internode using a cutter. Eachinternode sample was
scanned at four different scanning points from bottom to the
top (Fig. 1). Each section was imaginarily labelled as S1, S2, S3
andS4 following the sequence fromS1 to S4 (bottomto the top).
2.2. Reectance measurement
The reectance measurement method was chosen for this
study because this method is the easiest to perform as it re-
quires no contact with the fruit and light levels are relatively
high (Schaare & Fraser, 2000). Reectance measurement
methods have been successfully used to measure crops
quality by Chia et al. (2012) and Digmanand Shinners (2008). In
this study, a handheld Vis/NIR spectroradiometer (FieldSpec
HandHeld and FieldSpec Pro FR, 325e1075 nm, Analytical
Spectral Devices (ASD), Inc., Boulder, CO, USA) was used to
collect spectra data fromthe skin surface over the wavelength
range of 325e1075 nm in 1.5 nm intervals. The equipment
used has a spectral resolution (full width at half maximum
(FWHM) of a single emission line) of approximately 3 nm at
around 700 nm. The scanning was undertaken using the 25

eld-of-view (FOV) of the spectroradiometer. Two halogen

lamps (Lowell Pro-Lam 14.5 V tungsten bulb, Ushio Lighting,
Inc., Japan) were used to provide illumination to the sensor.
The lamps were placed at a distance of 800 mm and an angle
at 45

away from the sample. The distance between sample

and sensor was set at 70 mm, resulting in a measured spot of
0.031 mdiameter as calculated based on formula given by ASD
(2005). Distances between sensor and samples were main-
tained by xing the sensor to a tripod and the samples were
held by a xed sample holder.
The equipment was set to record the average of 20 scans
for each spectrum. All spectral data were stored in a computer
and processed using the RS3 software for Windows (Analytical
Spectral Devices, Boulder, CO, USA) designed with a graphical
user interface. Firstly, the reectance spectra (R) were trans-
formed into ASCII format by using the ASD ViewSpec Pro
software (Analytical Spectral Devices, Boulder, CO, USA).
Then, the absorbance spectra (A) were calculated as Log (1/R)
using R data. Four spectra from each internode (S1eS4) were
then averaged and later used for calibration against

Brix
values obtained by the refractometer method.
The experiment was conducted inside a light-proof mea-
surement box (900 600 400 mm). The box was built to
enclose the articial light source, detecting probes and the
samples from ambient light. Relative reectance spectra were
calculated by dividing stalk radiance with reference radiance
from a spectralon white reference panel for each wavelength.
In order to reduce signal-to-noise ratio, the rst and last 75 nm
data points were removed from the spectral data as recom-
mended by (Liu, He, &Wang, 2008). Therefore, only the regions
between 400 and 1000 nm were used for the calculations.
2.3.

Brix measurement
After the spectra acquisition, each internode was squeezed
using a clamp to extract juice samples for

Brix measurement.
Juice from each internode was collected in a small container,
properly shaken and poured onto handheld

Brix refractom-
eter (Model: RHB-32ATC, from Huake Instrument Co., Ltd,
Baoan, Shenzhen, China; the

Brix range is from 0 to 32% with
automatic temperature compensation) to measure the

Brix.
The spectra scanning and

Brix measurement were carried
out in the same day. Although the standard sugarcane quality
is measured using CCS which is derived from

Brix, Pol and
bre content, only

Brix value was used in this study because
it is a relevant industrial parameter for sugarcane industry
which is the easiest, least expensive quality parameter to be
measured with little preparation. The correlation between

Brix and CCS has been reported by Staunton, Donald, and

Pope (2011).
2.4. Pretreatment of spectral data
Both reectance and absorbance spectra were pre-processed
for optimal performance. Pretreatment of spectral data is a
key part of spectral analysis to improve the accuracy. Several
common chemometrics methods were investigated including
smoothing technique of moving average, multiplicative scat-
ter correction (MSC), rst and second derivatives, standard
normal variate (SNV) transformation and mean normal-
isation. After some trials and computations, MSCwas found to
be the best pretreatment method for this study. In fact, MSC
technique is the most popular normalisation technique
offered by most chemometrics software packages (Ns,
Isaksson, Fearn, & Davies, 2004). MSC is used to compensate
for additive (baseline shift) and multiplicative (tilt) effects in
the spectral data, which are induced by physical effects, such
as the non-uniformscattering throughout the spectrumas the
degree of scattering is dependent on the wavelength of the
radiation, the particle size and the refractive index. The Fig. 1 e Scanning positions along individual internodes.
b i o s ys t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 156
pretreatment processes were carried out using The Un-
scrambler V 9.6 software (Camo Process, AS, Oslo, Norway).
2.5. Principal component analysis (PCA)
PCA is a well-known multivariate data reduction method
which transforms original data into new orthogonal variables
referred to as principal components, or PCs (Purcell, Leonard,
OShea, & Kokot, 2005). PCA produces a reduced representa-
tion of the training data based on the maximum variation
between the spectra (Park et al., 2003). Usually, only a few PCs
or latent variables (LVs) are required to describe most of the
data variance with the rst PC accounting for the greatest
amount of variance. A low number of PCs were normally
desirable in order to avoid inclusion of signal noise in the
modelling (Xiaobo, Jiewen, Xingyi, & Yanxiao, 2007). In this
paper, PCA for Partial least square (PLS) models was imple-
mented using The Unscrambler V 9.6. During the building of
calibration models, PCA was also used to detect outlier sam-
ples which could affect model performance. In this paper, one
outlier was identied and removed before PLS modelling.
2.6. Partial least square
PLS is a well-known factor analysis multivariate method prin-
cipally applied for predictive purposes (Purcell et al., 2005).
It requires a calibration step in which a model is constructed
from a number of signicant factors, which are selected, for
example, by the well-known cross-validation leave-one-out
method. PLS analysis is widely utilized in near infrared spec-
troscopy analyses. PLS analysis can be performed to establish a
regression model to predict quality parameters of sugarcane.
PLS considers simultaneously the variable matrix Y (

Brix
values) and the variable matrix X (spectral data). Thus, PLS
method was used in this study to interpret the spectra and
develop both calibration and prediction models for sugarcane

Brix. Inthe development of the PLS model, full cross validation

(leave-one-out) was used to evaluate the quality and prevent
over tting of the calibration model (Arana, Jare n, & Arazuri,
2005). The maximum LVs number for an acceptable PLS model
isusuallyten(Moghimi et al., 2010). Inthispaper, thePLSmodels
were run using ten LVs as identied by The Unscrambler V 9.6.
External validation method was used in this paper to check
the PLS model performance. The external validation proce-
dure determines the predictive ability of an equation, based
on a sample set which has not been used in the calibration
development. Before the calibration, samples were divided
into two sets. One part (75% of samples) was used to develop a
prediction equation (calibration set) and another part (25% of
samples) was used to validate the predictive equation (vali-
dation set). Samples for validation were selected by taking one
of every four samples from the entire sample set, taking care
to ensure that each set included samples that covered the
entire range of

Brix values.
2.7. Description of classication model using the
articial neural networks
The ANN used in this paper was a perceptron model also
known as back-propagation perceptron. This type of network
was selected because it is an excellent pattern classier
(Torrecilla, Otero, & Sanz, 2004). The ANN selected was a feed-
forward network with supervised learning. The ANN model
consists of two layers with connections to the outside world
(an input layer where data are presented to the network and
an output layer which holds the networks response to given
inputs), and one hidden layers with ten neurons. More hidden
layers may cause over tting, since the network focuses
excessively on the idiosyncrasies of individual samples (Ruan,
Almaer, & Zhang, 1995). The input layer had ve neurons
which corresponded to rst ve PCs as determined by PCA
method.
After removing one outlier from the original training data
set (200), the PCA method was then run over the training set
consisting of 219 samples while the testing set had 71 sam-
ples. The output layer consists of ve neurons corresponding
to each

Brix range as dened in classication table (Table 1).
The ANNprocess is carried out by adjusting the weight of each
node, using the transfer function. The ANN algorithm
including PCA method for classication purposes in this study
was programmed and executed in Matlab (Version 7, The
Mathworks Inc. Natick, MA, USA). The transfer function
employed was the sigmoid function bounded between0 and 1.
The numerical values of the input and output variables used
by the ANN were normalised values in the range of 0e1.
3. Results and discussions
3.1. Overview of the spectra and statistic characteristics
of

Brix
The summary of statistical characteristics for calibration and
prediction data sets of stalk samples is shown in Table 2. The
calibration and prediction data set showed similar means,
ranges and standard deviation. Arelatively wide range of

Brix
values was obtained due to the inclusion of three different
varieties with different maturity stages. The range of

Brix
values for internode samples fromthe top to the bottomof the
Q155, Q208 and Q190 varieties were 7.6e22.2, 8e21.4 and 8e21,
respectively. The variation of

Brix values along the stalk for
each variety as measured from each internode sample is
shown in Fig. 2. The downward trend of

Brix values along the
stalk is consistent for all varieties. The graph also showed that
different varieties have different internode numbers. It can be
seen that the early maturing variety (Q155) had higher

Brix
values than other varieties especially for the several in-
ternodes at the bottom of the stalks.
The example of raw reectance spectra (R) and absorbance
spectra (A) of three samples having high (22

Brix), medium
Table 1 e Classication table for sugarcane

Brix.
Class

Brix range
High (H) 19.3e22.2
Medium high (MH) 16.5e19.2
Medium (M) 13.5e16.4
Medium low (ML) 10.5e13.4
Low (L) 7.6e10.4
b i os y s t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 157
(18

Brix) and low (14.2

Brix)

Brix values as measured by Vis/
SWNIR are shown in Fig. 3(a) and (c) respectively. In each
gure, no obvious difference could be seen in the shape of the
spectra for different

Brix values for both gures. However,
a baseline shift problem existing in raw reectance and
absorbance spectra has been eliminated using MSC method as
shown in Fig. 3(b) and (d) respectively.
3.2. The prediction of sugarcane

Brix
PLS models were developed using the pretreated reectance
spectra by MSC technique. The performance of the nal PLS
calibration models could be evaluated by the standard error of
calibration (SEC), the coefcient of determination for calibra-
tion (R
2
), the standard error of prediction (SEP), and the coef-
cient of determination for prediction (R
2
). A proper model
should have a low SEC, SEP and root mean square error of
prediction (RMSEP) and a high coefcient of determination for
both prediction and calibration models. The values of these
indices for PLS models developed using reectance and
absorbance spectra are shown in scatter plots in Fig. 4(a) and
(b) respectively. In both gures, the ordinate and abscissa
represent the predicted and measured values of the

Brix. The
R
2
and RMSEP values for reectance spectra were 0.91 and 1.41
while for absorbance spectra were 0.89 and 1.51 respectively.
The prediction performance of PLS models for both reec-
tance and absorbance spectra showed good agreement be-
tween the reference and estimated values. Shenk and
Westerhaus (1996) suggested that an R
2
value greater than
0.9 indicates excellent quantitative information of the models.
From Fig. 4, it can be seen that the PLS model developed
with reectance spectra performed better than the model
developed with absorbance spectra as indicated by R
2
and
RMSEP. This is reasonable nding since the absorbance spectra
is related to the presence of chemical components such as
sugars, while the reectance spectra of fruit contain informa-
tion on both the absorption as well as scattering properties of
the tissue (Nicola et al., 2008). Since this study was performed
on skin surface, scattering properties of the tissue would also
have contributed to the model performance. The ten LVs used
in PLS model were also found to be sufcient for predicting

Brix value from skin surface of sugarcane stalks.

The R
2
value of sugar content (

Brix) obtained in this paper

from reectance spectra was slightly better than those ob-
tained by Peirs, Lammertyn, Ooms, and Nicola (2001) with R
2
values ranged between 0.73 and 0.89 using different apple
varieties. The result of this study was also better than that
reported by Wu et al. (2008) who used Vis/NIR to detect an
infection on eggplant leaves with prediction ability with an
85% accuracy rate. It is also better than that reported by
Schaare and Fraser (2000) who obtained R
2
of 0.86 for pre-
dicting soluble solids content (SSC) on kiwifruit. Chia et al.
(2012) also reported lower prediction value (R
2
0.68) when
predicting sugar content from pineapple. This result is the
same as obtained by Khuriyati and Matsuoka (2004) for non-
destructively scanning tomato (R
2
0.91) on the skin to pre-
dict sugar content. However, the result of this study is lower
than the result (R
2
0.93) reported by Moghimi et al. (2010) for
predicting sugar content of kiwifruit. Overall, this proposed
technique has demonstrated that Vis/SWNIR can be used to
predict sugar content of sugarcane based on skin scanning.
This nding suggested that spectroscopic method has the
potential to be mounted on a chopper harvester for producing
a quality map across the eld.
3.3. Sugarcane

Brix classication by articial neural
network
Table 3 shows the accuracy of classifying sugarcane

Brix
using ANN with ve PCs. Five classes with their respective
threshold values were predened from the distribution of

Brix as discussed in Table 1. In this model, the accuracy of

classication ranged from 50% to 100% with the average ac-
curacy being 83.1%. The accuracy ranges obtained in this
study were better than that (45.2e93.5%) reported by Park et al.
(2003) for classifying soluble solids of Gala apples. However,
the result of this study was lower than the result (99.5% ac-
curacy) reported by Mohan, Karunakaran, Jayas, and White
(2005) who used ANN to classify grain quality.
From Table 3, it can be seen that two dominant classes
representing high and low

Brix level (H and L classes) had
achieved 100% accuracy. The M class was also good with an
accuracy of 90.9%. However, both inter-middle classes (MH
and ML) showed lower accuracy. This low accuracy (misclas-
sication) which only affected the boundary samples was
probably due to these two inter-middle classes had

Brix
values which were in a borderline between the main classes.
Thus, they are having similar reectance spectra with the
major classes. However, this misclassication is considered
minor because it is just affected the inter-middle group of
quality classes. The accuracy of the classication could be
improved if only three classes were used. Overall, the accu-
racy of this study can be considered as an acceptable and
useful especially for mapping the quality regions in the eld.
The ability to map the quality regions across the paddock
could help growers to better manage their elds thus improve
the crop production in both yield and quality.
10
12
14
16
18
20
22
24
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

B
r
i
x
Internode Number
Q155 Q208 Q190
Fig. 2 e Typical average

Brix values of internodes for
different sugarcane varieties.
Table 2 e Statistical characteristics of

Brix in sugarcane
stalks for calibration and validation.
Model Sample no Min Max Mean SD
Calibration 200 7.5 22.2 17.86 3.04
Prediction 71 8.2 22 17.83 2.93
SD Standard deviation.
b i o s ys t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 158
10
20
30
40
50
400 500 600 700 800 900 1000
R
e
f
l
e
c
t
a
n
c
e

(
%
)
Wavelength (nm)
22 Brix 18 Brix 14.2 Brix
(a) (b)
(d)
15
20
25
30
35
40
45
400 500 600 700 800 900 1000
R
e
f
l
e
c
t
a
n
c
e

(
%
)
Wavelength (nm)
22 Brix 18 Brix 14.2 Brix
2
3
4
5
6
7
8
400 500 600 700 800 900 1000
A
b
s
o
r
b
a
n
c
e

L
o
g

(
1
/
R
)
Wavelength (nm)
22 Brix 18 Brix 14.2 Brix (c)
2
3
4
5
6
7
400 500 600 700 800 900 1000
A
b
s
o
r
b
a
n
c
e

L
o
g

(
1
/
R
)
Wavelength (nm)
22 Brix 18 Brix 14.2 Brix
Fig. 3 e Typical sugarcane spectrum at different

Brix values for (a) raw reectance spectra; (b) raw reectance spectra
pretreated with MSC; (c) raw absorbance spectra; (d) absorbance data pretreated with MSC.
12
14
16
18
20
22
12 14 16 18 20 22
P
r
e
d
i
c
t
e
d

(

B
r
i
x
)
Reference (Brix)
R
RMSEP
n = 71
slope = 0.66
2
= 0.89
= 1.51
(a)
10
12
14
16
18
20
22
10 12 14 16 18 20 22
P
r
e
d
i
c
t
e
d

(

B
r
i
x
)
Reference (Brix)
(b)
R
2
n = 71
Slope = 0.72
= 0.91
RMSEP = 1.41
Fig. 4 e Prediction versus reference values for prediction models of sugarcane

Brix using (a) absorbance spectra and (b)
reectance spectra.
b i os y s t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 159
4. Conclusions
Vis/SWNIR is a portable, lowcost and non-destructive method
which was applied in this study to predict sugarcane

Brix
based on stalk scanning. The PLS models showed a good
agreement between the reference and estimated values for
both reectance and absorbance spectra with R
2
were 0.91 and
0.89 respectively. The ANN used to classify

Brix into several
quality classes had yielded good classication performance
ranging from 50 to 100% accuracy with overall accuracy of
83.1%. Overall, this study has demonstrated that the Vis/
SWNIR spectroscopy technique coupled with ANN has the
potential to be used for on-line quality measurement to full
the requirement of PA. This study also demonstrated that
quality prediction based skin scanning using a Vis/SWNIR
spectroscopy is feasible. This technique will be a better
alternative to current quality measurement which is based on
juice samples. Thus, the next study will focus on how to apply
this technique to map sugarcane

Brix during harvesting in
the eld. However, an effort is required to design a proper
sampling mechanism and installation conguration before
this method can be applied on a harvester. Spectral variations
due to different varieties, planting regions, seasons, weather
and cultural conditions also need to be considered in the
future works.
Acknowledgements
The authors acknowledge the nancial supports provided by
Ministry of Higher Education, Malaysia and National Center
for Engineering in Agriculture (NCEA), Toowoomba, Australia.
The authors also thank BSES Limited, Bundaberg, for
providing samples and equipment.
r e f e r e n c e s
Arana, I., Jare n, C., & Arazuri, S. (2005). Maturity, variety and
origin determination in white grapes (Vitis vinifera L.) using
NIRS. Journal of Near Infrared Spectroscopy, 13, 349e357.
ASD. (2005). Handheld spectroradiometer: Users guide version 4.05.
Suite A Boulder, USA: Analytical Spectral Devices, Inc.
Bramley, R. G. V. (2009). Lessons from nearly 20 years of precision
agriculture research, development, and adoption as a guide to
its appropriate application. Crop & Pasture Science, 60, 197e217.
Bramley, R. G. V., Panitz, J. H., Jensen, T., & Baillie, C. (2012).
Within block spatial variation in CCS e another potentially
important consideration in the application of precision
agriculture to sugarcane production. Proceedings of the
Australian Society of Sugar Cane Technologists, 34, 1e8.
Chia, K. S., Rahim, H. A., & Rahim, R. A. (2012). Prediction of
soluble solids content of pineapple via non-invasive low cost
visible and shortwave near infrared spectroscopy and articial
neural network. Biosystems Engineering, 113, 158e165.
Cox, G., Harris, H., Pax, R., & Dick, R. (1996). Monitoring cane yield
by measuring mass ow rate through the harvester.
Proceedings of the Australian Society of Sugar Cane Technologists,
152e157.
Digman, M. F., & Shinners, K. J. (2008). Real time moisture
measurement on a forage harvester using near infrared
reectance spectroscopy. Transactions of the ASABE, 51,
1801e1810.
Guthrie, J., & Walsh, K. (1997). Non-invasive assessment of
pineapple and mango fruit quality using near infra-red
spectroscopy. Australian Journal of Experimental Agriculture, 37,
253e263.
Haykin, S. (1994). Neural network: A comprehensive foundation. New
York, NY: Macmillan.
Johnson, R. M., & Richard, E. P., Jr. (2005). Sugarcane yield,
sugarcane quality, and soil variability in Louisiana. Agronomy
Journal, 97, 760e771.
Khuriyati, N., & Matsuoka, T. (2004). Near infrared transmittance
method for nondestructive determination of soluble solid
content in growing tomato fruits. Environmental Control in
Biology, 42, 217e223.
Lee, W. S., Alchanatis, V., Yang, C., Hirafuji, M., Moshou, D., &
Li, C. (2010). Sensing technologies for precision specialty crop
production. Computers and Electronics in Agriculture, 74, 2e33.
Liu, F., He, Y., & Wang, L. (2008). Comparison of calibrations for
the determination of soluble solids content and pH of rice
vinegars using visible and short-wave near infrared
spectroscopy. Analytica Chimica Acta, 610, 196e204.
Lu, R. (2004). Multispectral imaging for predicting rmness and
soluble solids content of apple fruit. Postharvest Biology and
Technology, 31, 147e157.
Magalha es, P. S. G., & Cerri, D. G. P. (2007). Yield monitoring of
sugar cane. Biosystems Engineering, 96(1), 1e6.
Menesatti, P., Antonucci, F., Pallottino, F., Roccuzzo, G.,
Allegra, M., Stagno, F., et al. (2010). Estimation of plant
nutritional status by Vis/NIR spectrophotometric analysis on
orange leaves [Citrus sinensis (L) Osbeck cv Tarocco]. Biosystems
Engineering, 105, 448e454.
Moghimi, A., Aghkhani, M. H., Sazgarnia, A., & Sarmad, M. (2010).
Vis/NIR spectroscopy and chemometrics for the prediction of
soluble solids content and acidity pH of kiwifruit. Biosystems
Engineering, 106(3), 295e302.
Mohan, L. A., Karunakaran, C., Jayas, D. S., & White, N. D. G.
(2005). Classication of bulk cereals using visible and NIR
reectance characteristics. Canadian Biosystems Engineering,
47(7), 7e14.
Montes, J. M., Utz, H. F., Schipprack, W., Kusterer, B.,
Muminovic, J., Paul, C., et al. (2006). Near-infrared
spectroscopy on combine harvesters to measure maize grain
dry matter content and quality parameters. Plant Breeding, 125,
591e595.
Nicola, B. M., Verlinden, B. E., Desmet, M., Saevels, S., Saeys, W.,
Theron, K., et al. (2008). Time-resolved and continuous wave
NIR reectance spectroscopy to predict soluble solids content
and rmness of pear. Postharvest Biology and Technology, 47,
68e74.
Ns, T., Isaksson, T., Fearn, T., & Davies, T. (2004). A user-friendly
guide to multivariate calibration and classication. Charlton,
Chichester, UK: NIR Publications.
Table 3 e ANN classication results.
Class

Brix
range
No. of test
sample in
each group
No. of correct
classication
Accuracy
(%)
High (H) 19.3e22.2 30 30 100
Medium
High (MH)
16.5e19.2 23 14 60.9
Medium (M) 13.5e16.4 11 10 90.9
Medium
Low (ML)
10.5e13.4 4 2 50
Low (L) 7.6e10.4 3 3 100
Total 71 59 83.1
b i o s ys t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 160
Panigrahi, S., & Hofman, V. (2003). On the go sugar sensor for
determining sugar content during harvesting. U.S. Patent No.
6,624,888.
Park, B., Abbott, J. A., Lee, K. J., Choi, C. H., & Choi, K. H. (2003).
Near-infrared diffuse reectance for quantitative and
qualitative measurement of soluble solids and rmness of
delicious and gala apples. Transactions of the ASAE, 46(6),
1721e1731.
Peirs, A., Lammertyn, J., Ooms, K., & Nicola, B. M. (2001).
Prediction of the optimal picking date of different apple
cultivars by means of VIS: NIR-spectroscopy. Postharvest
Biology and Technology, 21, 189e199.
Price, R. R., Johnson, R. M., Viator, R. P., Larsen, J., & Peters, A.
(2011). Fiber optic yield monitor for a sugarcane harvester.
Transactions of the ASABE, 54(1), 31e39.
Purcell, D. E., Leonard, G. J., OShea, M. G., & Kokot, S. (2005).
A chemometrics investigation of sugarcane plant properties
based on the molecular composition of epicuticular wax.
Chemometrics and Intelligent Laboratory Systems, 76, 135e147.
Qing, Z., Ji, B., & Zude, M. (2007). Predicting soluble solid content
and rmness in apple fruit by means of laser light
backscattering image analysis. Journal of Food Engineering, 82,
58e67.
Ruan, R., Almaer, S., & Zhang, J. (1995). Prediction of dough
rheological properties using neural networks. Cereal Chemistry,
72(3), 308e311.
Schaare, P. N., & Fraser, D. G. (2000). Comparison of reectance,
interactance and transmission modes of visible-near infrared
spectroscopy for measuring internal properties of kiwifruit
(Actinidia chinensis). Postharvest Biology and Technology, 20,
175e184.
Shenk, J. S., & Westerhaus, M. O. (1996). Calibration the ISI way. In
A. M. C. Davies, & P. C. Williams (Eds.), Near infrared
spectroscopy: The future waves (pp. 198e202). Chichester: NIR
Publications.
Staunton, S., Donald, D., & Pope, G. (2011). Estimating sugarcane
composition using ternary growth relationships. Proceedings of
the Australian Society of Sugar Cane Technologists, 33.
Torrecilla, J. S., Otero, L., & Sanz, P. D. (2004). A neural network
approach for thermal/pressure food processing. Journal of Food
Engineering, 62, 89e95.
Walsh, K. B., Guthrie, J. A., & Burney, J. W. (2000). Application of
commercially available, low-cost, miniaturised NIR
spectrometers to the assessment of the sugar content of intact
fruit. Australian Journal of Plant Physiology, 27, 1175e1186.
Wendte, K. W., Skotnikov, A., & Thomas, K. K. (2001). Sugar cane
yield monitor. U.S. Patent No. 6,272,819.
Williams, P. C., & Norris, K. H. (2001). Near-infrared technology in the
agricultural and food industries (2nd ed.). St. Paul, MN: AACC.
Wu, D., Feng, L., Zhang, C., & He, Y. (2008). Early detection of
Botrytis cinerea on eggplant leaves based on visible and near-
infrared spectroscopy. Transactions of the ASABE, 51(3),
1133e1139.
Xiaobo, Z., Jiewen, Z., Xingyi, H., & Yanxiao, L. (2007). Use of FT-
NIR spectrometry in non-invasive measurements of soluble
solid contents SSC of Fuji apple based on different PLS
models. Chemometrics and Intelligent Laboratory Systems, 87,
43e51.
Xing, J., Symons, S., Shahin, M., & Hatcher, D. (2010). Detection of
sprout damage in Canada Western Red Spring wheat with
multiple wavebands using visible/near-infrared hyperspectral
imaging. Biosystems Engineering, 106, 188e194.
b i os y s t e ms e ng i ne e r i ng 1 1 5 ( 2 0 1 3 ) 1 5 4 e1 6 1 161