Quantitative Determination of Sugar Cane Sucrose by Multidimensional Statistical Analysis of their Mid-Infrared Attenuated Total Reflectance Spectra*
ABSTRACT A fast and accurate method for determining the sucrose content of sugar cane juice has been developed. The application of principal component regression (PCR) has been proposed for the development of a prediction equation of sucrose content by mid-infrared spectroscopy. An attenuated total reflectance (ÀTR) cell is used in place of the more familiar hans-mission cell. PCR involves two steps: (1) the creation of new synthetic variables by principal component analysis (PCA) of spectral data, and (2) multiple linear regression (MLR) with these new variables. Results obtained by this procedure have been compared with those obtained by the conventional application of polarization.
- [Show abstract] [Hide abstract]
ABSTRACT: Complex-formation between carbohydrates and cations may have important biological implications. Potassium ions and sucrose interactions have been observed by the examination of the spectral patterns obtained by multi-dimensionnal analysis (Principal Component Analysis) applied to infrared spectra of biological samples. Several authors have extensively studied ions-sugars interactions particularly calcium and magnesium-sugar interactions and have shown that these interactions induced shifts and splitting in infrared absorption bands.We here report to use Principal Component Analysis and Principal Component Regression for the quantitative determination of K ions in biological solutions by investigating interactions between this cation and sucrose molecules. The spectral pattern that describes the principal component (axis 1), features absorption peaks at 925, 997, 1053, 1112 and 1136 cm-1 that are characteristic of sucrose. The correlation coefficient between axis 1, representative of pure sucrose, and the values of potassium concentrations is 0. 5213. The spectral pattern of the second axis is also associated with the sucrose pattern. The shifts and splittings of the absorption bands that are observed in this second spectral pattern are associated with potassium-sucrose interactions. The correlation coefficient between this second axis (axis 3) and the K concentration is equal to 0. 2152. Most of the information relative to potassium ions are contained in this axis. These informations appear as interactions between sucrose and the cation. With the first two axes, the correlation coefficient reaches 0. 7365 and this opens the possibility of quantifying K in biological samples. The other axes improved the correlation coefficient, which reached 0.9945 with the first ten axes. For the predicted concentrations of K, the bias and standard deviation values, 8 × 10-2 and 0.43 respectively, showed that the predicted values are very close to those determined by flame photometry.Hence, even if K cation does not have a Mid-infrared spectral fingerprint, interactions between K and sucrose in raw sugar cane juices allow the indirect quantitative determination of this ion.Spectroscopy Letters 01/1997; 30(1):1-16. · 0.67 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The Fourier transform infrared (FT–IR) spectra of palm oil samples, in the range between 3025 and 2992 cm−1, were used to compare different multivariate calibration techniques for quantitative iodine value (IV) determination. Forty-two spectra of palm oil with IV ranging between 53 and 65 were used to create calibration models based on partial least squares (PLS) and principle component regression (PCR) methods using different baseline types. The methods were compared with respect to the number of factors, coefficient of determination (R2) and accuracy of estimation. The standard error of prediction (SEP) ratios were calculated to compare the prediction capabilities of these calibration methods. The calibration models generated the number of factors from 3 to 7, R2 of 0.94443 to 0.98853, standard error of estimation (SEE) of 0.32 to 0.69 and SEP ratios of 1.43 to 13.48.Food Chemistry 01/1999; 67(2):193-198. · 3.33 Impact Factor
Quantitative Determination of Sugar Cane Sucrose by
Multidimensional Statistical Analysis of their
Mid-Infrared Attenuated Total Reflectance Spectra*
rnBoÉNTC CADET,T DOMINIQUE BERTRAND, PAUL ROBERT, JOSEPH MAILLOT,
JULES DIEUDONNÉ, ANd CLAUDE ROUCH
(Jniuersite de la Réunion, Laboratoire de Chitnie Organique, Faculté des Sciences, BP 5,97490 Sainte Clotilde, Ile de la Réunion,
France (F.C., C.R.); INRA, Laboratoire de Technologie Appliquée à la Nutrition, Rue de la Géraudière, BP 527,44026 Nantes
Céd.ex, France (D.8., P.R.); and Centre Technique Interprofessionnel de la Canne et du Sucre, Laboratoire de Recherche et
Déueloppernent, Route de I'ONF, B.P. 140,97464 Saint Denis Cédex, Ile de la Réunion, France (J.M., J.D.)
A fast and accurate method for determining the sucrose content of sugar
cane juice has been developed. The application of principal component
regression (PCR) has been proposed for the development of a prediction
equation of sucrose content by mid-infrared spectroscopy. An attenuated
total reflectance (ÀTR) cell is used in place of the more familiar hans-
mission cell. PCR involves two steps: (1) the creation of new synthetic
variables by principal component analysis (PCA) of spectral data, and
(2) multiple linear regression (MLR) with these new variables. Results
obtained by this procedure have been compared with those obtained by
the conventional application of polarization.
Index Headings: Mid-infrared spectroscopy; ATR; Prediction; Sucrose.
Polarimetric analysis is the method traditionally em-
ployed to determine the sucrose content of juice samples
in the cane sugar industry. Although this technique is
adequate under ideal conditions, it may have serious
limitations under circumstances in which optically active
substances other than sucrose are present.l Under such
circumstances more accurate methods of sucrose analysis
are required.2,3 These include enzymatic methods, gas-
liquid chromatography with prior derivatization, paper
chromatography, and liquid chromatography. However,
considerations of cost, time of analysis per sample, and
complexities involved in these technique have hampered
the acceptance ofthese methods by cane sugar industries.
Near-infrared reflectance is widely used to rapidly
measure the composition of foods and food products,a
notably cereal products.s
Near-infrared reflectance spectroscopy has already
been used for the analysis of sugar content in sugar cane
samples. The values obtained are comparable to those
measured with a polarimeter.6 The fundamental vibra-
tion-rotation bands and the skeletal vibrations fall in the
mid-infrared range (400-4600 cm-'). However, use of the
mid-infrared to characterize complex products such as
foods and food products has long been precluded by
technical problems. The advent of Fourier transform
spectrometers, with their many advantages over tradi-
tional dispersive instruments, has improved the quality
of infrared data while making data collection faster and
Received 23 May 1990; revision received 1 August 1990.
* This publication is dedicated to Professor Jean-Claude Meunier.
t Author to whom correspondence should be sent.
166 Volume 45, Number 2, 1991
In addition, new devices and easier techniques have
been developed to study opaque powders.8 Mid-infrared
spectra contain a wealth of information about material
constituents. Complex spectra make interpretation dif-
ficult, but this problem may be overcome by using mul-
tidimensional statistical analyses.e'lo
This study investigates mid-infrared ATR potential-
ities to determine the amount of sucrose present in sugar
cane juice, using new technical developments combined
with multidimensional statistical analyses. To our
knowledge, this combination has not yet been used for
the analysis of sugar cane juice and could prove to be
MATERIALS AND METHODS
Samples. Sampling of sugar cane by coring is used.
The average core sample is about 7000 g. After pulver-
ization, a subsample of approximately 1000 g is removed.
A hydraulic press is used to extract juice from the sub-
samples obtained from the coring sampler and from the
The subsample is pressed for two and a half minutes
at 250 bars. A juice preservative is added (lead subacetate
at 2.5 g for 150 mL), and storage is carried out at 4oC.
In reconstituted sugar cane juices, sucrose, glucose,
and fructose are mixed in proportions in accordance with
those observed in natural sugar cane juices analyzed in
our laboratory. Natural juices with sucrose concentra-
tions ranging from I to 23% are used as models. The
glucose and fructose concentrations are those corre-
sponding to these natural juices.
Glucose, Reducing Sugars, and Sucrose Determina-
tions. The enzymatic and colorimetric determinations of
glucose and reducing sugars, respectively, have been de-
termined by a continuous flow technique (the IRIS-TDF
flow cell analysis method). Glucose oxidase and peroxi-
dase are used for the specific determination of glucose.
The resulting product (monoimino-p-benzaquinone-4-
phenazone) is determined spectrophotometrically at 505-
520 nm. The reducing sugars are determined by spec-
trophotometry at 460 nm after reaction with neocuproine
hydrochloride and cupric sulfate.
The accuracy and reproducibility of this approach are
very high and agree with the demands required for sugar
cane analysis. The experimental analysis error (differ-
ence between the calculated and known concentrations
ooos-? 028 /91 I 4502-0166S2.00/0
@ 1991 Societÿ for Applied Spectroscopy
Enzymatic and polarimetric measures
Polarization corrected Variation
polarization (b - a\
Pure sucrose solutions
(d - b)
(c - b)
in an analyüical sample) is less than 2% for both glucose
and reducing sugars.
Sucrose is determined either by direct polarimetric
measurement (POL) or by correction from the effect of
glucose and fructose. The correction is given by:
(sucrose) : POL - 0.7921(glucose) + 1.38885(fructose)
(fructose): (reducing sugar) - (glucose)
(sucrose) : POL - 2.18095(glucose) + 1.38885(reducing
where: POL : the polarimetric measure (noncorrected
sucrose content); (glucose) : the enzymatic determina-
tion; (reducing sugars) : the colorimetric determination;
and 0.7921 and 1.38885 : the angular rotation measure-
ments for glucose and fructose, respectively.
Mid-Infrared Attenuated Total Reflectance Spectra.
Mid-Fourier Transformed Infrared (Mid-FT-IR) spectra
APPLIED SPECTROSCOPY 167
Frc. 1a. Fourier transformed mid-IR spectrum of sugar cane juice.
were collected on a Michelson-100 Fourier transform
spectrophotometer. Attenuated total reflectance spectra
were obtained with a Specac Overhead ATR system. The
crystal of the reflectance element is made from zinc sele-
nide, a material that is quite inert to water; it is rapidly
cleaned between samples by being sprayed with water
and then dried with filter paper.
The data were recorded from 800 to 1250 cm-1in 4-cm
increments at log(l/R), in which È is the ratio of the
reflected intensity for the background to that of the sam-
ple. Although the ATR experiment does involve the re-
flection of the radiation within a crystal, the interaction
of the radiation with the sample is a transmittance of
radiation through the sample; this depth of penetration
is wavelength dependent, but it is still passing through
a finite layer of the sample. For this reason, plots can
read according to absorbance (or transmittance). The
combination of four scans resulted in an averâge spec-
trum. The intensity of the spectra was low; the highest
peaks had log(1/À) values lower than 0.60 on baseline
Mathematical Treatments. Mathematical treatment
was performed on a Nec personal computer with software
Frc. 1b. Fourier transformed mid-IR spectra of: (O) HrO; (tr) fructose;
(O) glucose, (*) sucrose.
168 Volume 45, Number 2, 1991
Frc. 2. Sucrose content in sugar cane juices: reference concentrations
(corrected polarization) vs. predicted concentrations. (a) Prediction set;
(b) verification set.
written and developed at the Laboratoire de Technologie
des Aliments des Animaux, INRA, Nantes.
Multidimensional statistical analyses, such as princi-
pal component analyses (PCA), describe variation in
multidimensional data by a few synthetic variables. These
synthetic variables are linear combinations of all the orig-
inal variables and have the advantage of having no cor-
relation with each other. Simpler descriptions of data
sets are thus obtained with minimal loss of information.
These treatments were used for morphological analysis
of spectralr and for graphical representation of spectra
Principal component regression (PCR) was used to
Predtcted cotcentrât lons
a 5 x10
Dj.fference between predicted sucrose (IIID-FTIR) ancl sucrose content
(corrected polarization) .
Frc. 3. Frequency ofdistribution vs. deviation between sucrose contents (corrected polarization) and predicted concentrations for natural sugar
establish a prediction equation. PCR is basically a mul-
tilinear regression applied to scores assessed by PQ4.'z'ts
Interest in the introduction of scores according to their
predictive ability had already been shown.ra'15
For the prediction equation, the number of regression
terms was predetermined by specifying that the last term
introduced would have a probability of being nonsignif-
icant, its value being determined as less than 0.05 (ac-
cording to the .F' test). PCA was applied to the spectra
from 800 to L250 cm-l (with 235 data points used as
principal variables). Spectra were centered prior to PCA
where X,, : centered data; 4,, : spectral data (lo9 t/R)
of spectrum i and wavelength j; A, : mean value of
TABLE II. Influence of the number of regression terms in PtCÀ prediction equation on the precision of the results (sucrose content prediction).
Set of calibration
Number of introduced
terms PCA variable
Set of verification
Mean deüation deviation
spectral data at wavelength j for every spectrum; A, :
mean value of spectral data of spectrum i for every rtrave-
length; and A : average mean of all spectral data in the
collection. The resulting table therefore has a sum of
every row or every column equal to 0.
The relevance of the prediction equations was estab-
lished by the assessment of the bias and of the root mean
square of difference (RMSD) according to Kruschka.l6
Three prediction equations were achieved by using
PCR and changing the calibration set. One was estab-
lished on pure sucrose solutions ranging from 1 to25%
(g/100 mL). The second one used mixtures of sugars in
accordance with the observed proportions in sugar cane
juice (from 10.39 to 20.23% for sucrose). The last one
used the calibration set ofnatural sugar canejuices (from
9.47 to L9.93%).
170 Volume 45, Number 2, 1991
RESULTS AND DISCUSSION
Chemical Yalues and Spectra. Chemical values of the
verification set are given in Table I. Sucrose ranged from
11.78 to 20.91 9/100 mL with an average value of 16.43
and a standard deviation equal to 1.80. Glucose and re-
ducing sugars, respectively, ranged from 0.07 to 0.56 (av-
erage value :0.24, standard deviation : 0.13) and 0.11
to 1.09 (average value : 0.49, standard deviation -- 0.26).
The results of sucrose content estimation by polariza-
tion without correction were compared with those as-
sessed by taking into account the effect of other sugars.
The bias and the RMSD were, respectively, 0.163 9/100
mL and 0.087 9/100 mL. The bias was significantly dif-
ferent from 0 (P > 0.05).
The effects of glucose and reducing sugars on the po-
larimetric measurement were therefore not negligible.
Frc. 4. Principal component analysis, Discrimination ef gemples of calibration according to sucrose content (g/100 mL).
Fre. 5. Morphological analysis: §pectral pattern ofprincipal component I (97.L8% oftotal variance). X values are wavenumbers (from 1250 to
800 cm-l) and Y vàues are iactorial coorAinates for the represented principal component (arbitrary intænsity units).
The optically active substances result in an undervalu-
ation of sucrose content, which is often used for deter-
mining the cost of cane sugar in the sugar cane industry.
Figure la shows a typical spectrum of sugar cane juice.
The most relevant area for sugar identification and pre-
diction is from 1250 to 800 cm-1. In natural products
such as sugar cane juice and other heterogenous mate-
rials, the various constituents often have overlapping,
coinciding, or mutually interfering absorption bands. This
region can be regarded as being the result of the super-
positioning of the spectra of individual components,
principally glucose, fructose, and sucrose.
The spectra of solutions of these pure sugars in the
runge L250 to 900 cm-1 are given in Fig. 1b.
Reflectance data plotted as log(1/8) vs. wavelength
produce a curve comparable to an absorption spectrum
and peaks occur at wavelengths corresponding to the
absorption bands of each sugar sample. Spectra of su-
crose, glucose, and fructose show some similarities.
Sucrose mainly differs in the absorption band at 996.7
cm-1. Sucrose is known to be a disaccharide made up of
a-D - glucopyranosyl and 0-D -fructofuranosyl groups; this
band is generally associated with the disaccharide link.
As the absorption bands are overlapping, it was im-
possible to measure the concentration of individual sug-
ars by direct application of the Beer-Lambert law on
separated relevant peaks.
Predictions by Using PCR on Spectra. Table I shows
the results obtained with reflectance measurements and
When calibration standards are pure sucrose solutions,
the coefficient of correlation is 0.9999 and the RMSD is
3.4.10-'z. The application of this equation to sugar cane
juice gave poor results (bias : 1.73, RMSD : 0.33).
Sugar cane juice is a very complex mixture comprised
of sugars, organic acids, and amino acids, and there are
certainly significant interactions between these compo-
nents; and the spectrum of pure sucrose can be quite
different from that of sucrose in a mixture.
Another prediction equation was established with re-
constituted juice, in which sucrose, glucose, and fructose
were added to water in the szlme concentrations that were
observed in the in uiuo sugar cane juices. The coefficient
of correlation was equal to 0.9998; the RMSD values of
the prediction set and verification set \ryere, respectively,
0.56 and 0.37. Utilization of reconstituted juice therefore
improved the prediction.
The last equation, established on natural sugar cane
juices, gave an RMSD of 1.7'10*2 and 0.12, respectively,
for the prediction and the verification sets.
The bias was very low ( - 3.05 ' 10-'z). The results given
in Fig. 2a and 2b compare favorably with those obtained
Figure 3 shows the distribution of the residuals (su-
crose observed minus sucrose prediction) for the verifi-
cation set (natural sugar cane juices). The histogram
seemed rather well balanced, with no outliers.
Table II shows the influence of the first ten regression
terms in the PCR equation on the precision of the result
(sucrose content prediction).
APPLIED SPECTROSCOPY 171
The first introduced component gave a correlation co-
efficient equal to 0.9718. Other components improved
Principal Component Analysis. Figure 4 shows the most
predictive map (components L and 2). Samples are dis-
tributed according to their sucrose content. Axis l- seems
to be representative of sucrose concentration. The spec-
tral pattern (eigenvectors) of component 1 (Fie. 5) is
in agreement with this distribution. Spectral pattern 1
shows positive bands representative of the disaccharide
link of saccharose (997 cm-') and the ring C-O stretching
vibrations (1054, 1105 cm-l); the absorption band atLl}7
is related to the strongly coupled sugar COH bending
These bands were already visible by direct examina-
tion of the original spectra, although with slight shifting.
Attenuated total reflectance mid-infrared Fourier
transform spectroscopy shows potential for the quanti-
fication of sucrose in clarified sugar cane juice.
In sugar cane juice, significant wavenumbers have been
identified for sucrose (927.59; 997.02; L054.87; 1116.51;
1137.80 cm-t).Prediction equations give results that are
more accurate and, in particular, more pertinent than
those obtained by polarization. The use of PCA variables
requires only a short computation time. The whole pro-
cedure (PCA + multiple linear regression * prediction
with about 59 samples) requires no more than twenty
minutes. The spectral acquisition time is shorter than
the time required for polarization measurements. These
results could prove to be interesting for the sugar cane
This work was supported by a grant from the Ministère de la Re-
cherche et de la Technologie and the Conseil Gènéral.
1. G. P. Meade and J. C. P. Chen, Cane Sugar Handboo& (J. Wiley
and Sons, New York, 1985), 11th ed.
2. M. A. Brokensha, R. H. Niemeyer, and K. J. SchÊifler, "A Com-
parison of the Estimation of Sucrose in Sugar Cane Juice by
Polarimetric and Gas Liquid Chromatographic Methods," in Pro-
ceedings of the 52nd Congress of the South African Sugar Tech'
nologists' Association (South African Sugar Technologists' Asso-
ciation, Mount Edgecombe, Natal, 1978), pp. 54-58.
3. K. J. Schâfier and I. A. Smith, "True Sucrose versus POL: The
Effect on Cane Quality and Factory Balance Data," in Proceedings
of the 52nd Congress of the South African Sugar Technologists'
Associotion (South African Sugar Technologists' Association, Mount
Edgecombe, Natal, 1978), pp. 59-63.
4. K. H. Norris, "Near-Infrared Reflectance Spectroscopy," inCereals
78: Better Nutrition for the World's Millions, Y. Pomerantz, Ed.
(American Association of Cereal Chemists, Manhattan, Kansas,
1978), p. 245-254.
5. B. G. Osborne, S. Douglas, and T. Fearn, J' Food Technol. 17,355
6. C. B. Suerzut, A. French, S. Faught, and L. R. Vergma, "Sugar
Cane Analysis Using NIR Spectroscopy," presented at the South-
west Region Meeting of the American Society of Agricultural En-
?. P. R. Griffiths and J. A. de Haseth, Fourier Transform Infrared
Spectrometry (J. Wiley and Sons, New York, 1986).
8. M. P. Fuller and P. R. Griffiths, Anal. Chem. 50, 1906 (1978).
9. M. F. Devaux, D. Bertrand, and G. Martin, Cereal Chem. 63, 151
10. I. A. Cowe and J. W. McNicol, Appl. Spectrosc. 39, 257 (1985).
11. J. le Nouvel,Etude d'une Famille de Courbes par Méthodes d'Ana-
lyse de Données . Application d I' Analyse Morphologique de Courbes
Prouenant de Données Médicales, Thèse de 3 ème cycle, Université
Rennes I, France (1981).
12. J. Lefebvre, Introduction aux Analyses Statistiques Multidirnen-
sionnelles (Masson, Paris, 1983), 3rd ed., pp. 137-148.
13. L. Lebart, A. Morineau, and N. Tabard, Techniques de la Descrip-
tion Statistique (Dunod, Paris, 1977), pp. 7-46.
14. P. Dagneli e, Analy se Statistique d Plusieurs Variables (Les Press-
es Agronomiques de Gembloux, Belgium, 1975), pp. 185-190.
15. D. Bertrand, M. Lila, V. Furtoss, P. Robert, and G. Downey, J. Sci.
Food Agric. 41,299 (L987).
16. W, R. Hruschka, "Data Analysis: Wavelength Selection Methods,"
ln Near-Infrared Technology in the Agricultural and Foods In-
d.ustries, P. Williams and K. Norris, Eds. (American Association
of Cereal Chemists, St. PauI, Minnesota, 1987), pp. 35-55.
17. M. Mathlouthi and C. Luu, Carbohydr. Res. 81, 203 (1980).
18. T. W. Barrett, Spectrochim. Acta 37À 233 (1981).
172 Volume 45, Number 2, 1991