J. Agric. Food Chem.

2006, 54, 7437−7443 7437

Measurement of Soluble Solids Contents and pH in Orange

Juice Using Chemometrics and Vis−NIRS
HAIYAN CEN, YONG HE,* AND MIN HUANG
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China

The potential of visible and near-infrared reflectance spectroscopy (vis-NIRS) was investigated for
its ability to nondestructively detect soluble solids contents (SSC) and pH in orange juices. A total of
104 orange juice samples were used for vis-NIRS at 325-1075 nm using a field spectroradiometer.
Wavelet packet transform, standard normal variate transformation (SNV), and Savitzky-Golay first-
derivative transformation were applied for the preprocessing of spectral data. The chemometrics of
partial least-squares (PLS) regression analysis was performed on the processed spectral data. The
evaluation of SSC and pH in orange juices by PLS regression with SNV showed the highest accuracy
of the three preprocessing methods. The correlation coefficient (r), standard error of prediction, and
the root-mean-square error of prediction for SSC were 0.98, 0.68, and 0.73, respectively, whereas
those values for pH were 0.96, 0.06, and 0.06, respectively. The “fingerprint” representing features
of orange juices or reflecting sensitivity to some elements at a certain band was proposed on the
basis of regression coefficients. It is very useful in the field of food chemistry and further research on
other materials. It is concluded that the vis-NIRS technique combined with chemometrics is promising
for the fast and nondestructive detection of chemical components in orange juices or other materials.

KEYWORDS: Orange juice; vis-NIR; SSC; pH; preprocessing; PLS

INTRODUCTION appears. All of these factors have emphasized the need for
reliable techniques to authenticate orange juices.
With the development of the food industry, the problems of
food additives and safety have attracted much attention, There are many methods to determine the internal quality of
especially for beverages. Orange juice is considered to be one orange juices reported in the literature. Common detecting
of the most important and common beverages in people’s daily techniques used in the field of food chemistry are fluorescence
life. Its quality is defined by physical characteristics (fluid, color, analysis, chromatographic determination, photospectrometric
and odor) and chemical parameters [soluble solids content analysis, electroanalysis, and capillary electrophoresis (3, 4).
(SSC), pH, carbohydrates, and vitamins], which determine the Hu et al. (5) used the fluorescence technique to detect adultera-
taste and nutrition of orange juices. tion of citrus juice with the emission wavelength region changing
SSC and pH are two significant properties in orange juices between 284.6 and 296.5 nm. A chemiluminescence method
that affect the consumer’s appreciation for the selection of for the determination of citric acid using tris(2,2′-bipyridine)
beverages. The SSC (1) in the fruit juice is mainly sugars, such ruthenium was also developed by some researchers (6). Saavedra
as fructose, sucrose, and glucose, and pure fruit juice contains et al. (2) developed and validated a capillary electrophoresis
∼9-15% of SSC. In orange juice, acids are also important method for the direct measurement of organic acids in fruit juice.
sources of nutrition and freshness with different kinds of organic Scordino and his partners (7) applied high-performance liquid
acids, including citric, tartaric, and malic acids (2). Because chromatography (HPLC), ultraviolet (UV), and mass spectra
organic acids have different abilities to dissociate hydrogen ion, to test and analyze the quality of orange juices. Most of these
the pH calculated by -log [H+] was used to show the acidity methods are based on the complex processing for samples, and
in orange juice. Also, the pH plays an important role in the the use of chemical reagents is also a problem if the economic
food processing industry related to the variation of color, benefit and safety are considered in the experiment. All of these
microbial control, taste, and authentication of food. Because a methods are categorized as destructive, and they are not very
great deal of nutrient elements will be lost in the processing of simple or convenient in practical application. For some tradi-
fresh orange juices, sugars, acids, water, and other additives tional testing methods, it takes a long time to obtain the testing
will be added into the juice to adjust the taste and color, which result for one sample from preparation to the end (8). Thus, a
is where the problem of adulteration of beverages consequently rapid, safe, and nondestructive method is needed to qualify for
testing the internal quality of orange juices.
* Author to whom correspondence should be addressed (telephone Recently, research has been focused on the development of
+86-571-86971143; fax +86-571-86971143; e-mail yhe@zju.edu.cn). a vis-NIRS technique for its potential application in the field
10.1021/jf061689f CCC: $33.50 © 2006 American Chemical Society
Published on Web 09/02/2006
7438 J. Agric. Food Chem., Vol. 54, No. 20, 2006 Cen et al.

Table 1. Brands of Orange Juice in Experiment

SSC pH
brand class production company range mean range mean
Minute Maid beverage Coca-Cola 2.5−11.75 5.42 3.46−3.75 3.60
Masterkong Day C beverage Masterkong 2.00−11.50 5.45 3.30−3.53 3.41
Qoo beverage Coca-Cola 2.75−13.75 6.36 3.54−3.76 3.66
Nongfu Orchard pure Nongfuspring 2.50−12.00 5.60 3.95−4.08 4.02
Masterkong beverage Masterkong 2.00−12.25 5.57 3.37−3.57 3.47
Unif beverage Unif 2.00−11.00 4.90 3.56−3.71 3.63
Great Lakes pure Great Lakes 2.00−13.00 6.36 3.91−4.29 4.00
Huiyuan pure Huiyuan 3.25−12.00 7.79 3.90−4.02 3.94

of food chemistry. The advantage of vis-NIRS is that not only The reference methods for determination of SSC and pH were chosen
can chemical structures be assessed through the analysis of the by considering the precision and convenience of measurement. Samples
molecular bonds in the visible and near-infrared reflectance were taken for SSC measurement by using an Abbe benchtop
spectrum but also a characteristic spectrum that represents the refractometer (2WAJ 0-95 °Brix, Shanghai, China), with a 0.02 °Brix
accuracy using temperature correction. A pH-meter (SJ-4A, Instrument
“fingerprint” of the sample can be built (9). He et al. (10)
Co. Ltd., Shanghai, China) with an accuracy of 0.01 was used to
analyzed apple brands using NIRS with principal component measure the pH.
analysis (PCA) and a BP neural network (BPNN) model.
Reflectance Measurement. Orange juice samples were poured about
Tsuchikawa and Hamada (11) proposed time-of-flight near- half full into glass sample containers of 120 mm diameter and 10 mm
infrared spectroscopy to detect sugar and acid contents in apples, height. All samples were prepared for vis-NIRS analysis at 325-1075
and three chemometric models were built, including partial least- nm. For each sample, three reflecting spectra were taken for three
squares regression (PLSR), principal component regression equidistant rotation positions of ∼120° around the container center with
(PCR), and multiple linear regression (MLR). In other papers, a field spectroradiometer [FieldSpec HandHeld (HH), VNIR (325-
nondestructive determination of solids and carotenoids in tomato 1075 nm)], Analytical Spectral Devices (ASD), Inc., Boulder, CO],
products by near-infrared spectroscopy was also presented (12). using RS2 V4.02 software for Windows designed with a graphical user
Meanwhile, classification and adulteration of beverages were interface (GUI) from ASD. The instrument uses a sensitivity 512-
studied by some researchers with NIRS (13). Nevertheless, SSC element, photodiode array spectroradiometer, with the resolution of 3.5
nm. The scan number for each spectrum was set to 10 at the same
and pH in orange juices were little calibrated using vis-NIRS.
position; thus, a total of 30 individual spectra were properly stored for
Due to the hidden information in spectral data, at present, later analysis. Considering its 25° field-of-view (FOV), the spectro-
particular attention has been paid to the data mining of numerous radiometer was placed at a height of ∼100 mm and 45° angle away
spectral data with chemometrics. PLSR has been proved to be from the center of the sample container. A light source of Lowell pro-
a more excellent method than the others, such as MLR and PCR lam 14.5 V bulb/128690 tungsten halogen was placed ∼300 mm away
(11, 14, 15). In the analysis of spectra, there are large baseline from the viewing area. To achieve the relative reflectance measure-
shifts and noises in the spectra with a broad wavelength region. ments, the white reference (a white panel purchased with the spectro-
Thus, the selection of suitable preprocessing methods is also radiometer used as white reference) was collected before scanning
an important step in the process of spectral analysis. Dou et al. samples to obtain a nice, clean, 100% reference line.
used four different preprocessing methods [first derivative, Preprocessing of Spectral Data. Due to the potential system
second derivative, standard normal variate transformation imperfection, obvious scattering noises could be observed at the
(SNV), and multiplicative scatter correction (MSC)] to process beginning and end of the spectral data. Thus, the first and last 75
wavelength data were eliminated to improve the measurement accuracy,
NIR spectra of compound aminopyrine phenacetin tablets, and
that is, all visible and NIR spectroscopy analysis was based on 400-
the SNV preprocessing spectra were found to provide the best 1000 nm. The above spectral data preprocessing was finished in
results (16). Also, some researchers applied wavelet packet ViewSpec Pro V4.02 (Analytical Spectral Devices, Inc.). Absorbance
transform (WPT) in pattern recognition of NIRS data (17). for the scan was recorded as log [1/R], and all spectral records were
The objective of this study was to assess the potential of vis- checked visually and averaged. The absorbance wavebands were then
NIRS as a rapid and nondestructive technique to detect SSC preprocessed using WPT, SNV, and S. Golay first-der, respectively.
and pH in orange juices depending on the multivariate calibra- Wavelet transform (WT) is a very popular kind of operation today
tion PLS with different preprocessing methods [WPT, SNV, due to its application in chemometrics and signal processing (18, 19),
and Savitzky-Golay first-derivative (S. Golay first-der)]. whereas WPT (20) is a derivative of WT that has many advantages in
“Fingerprint” analysis based on regression coefficients was also the extraction of information from the certain time and frequency. In
proposed so that further research on other materials would be this experiment, WPT was proposed as a preprocessing method for
more feasible. denoising. The wavelet function symlets 6 was adopted to decompose
the wavelet packet into five layers, wpbmpen was used to determine
the threshold, and noises in the spectra were eliminated via the function
MATERIALS AND METHODS wpdencmp (21). WPT was performed in software MATLAB 7.01 with
the edited program.
Sample Preparation and Reference Method. Eight brands of
orange juices purchased from the local market were selected for the SNV removes the multiplicative interferences of scatter, particle size,
vis-NIRS analysis (shown in Table 1), and each brand includes a and the change of path length (22). For an individual spectrum to be
certain number of samples produced at different dates. For analytical processed by SNV, it is calculated as
purposes, the samples were also diluted. Finally, 104 samples were

x
obtained in this experiment. They were placed in airtight glass bottles, m
stored in an ice-filled cooler, and transported to the laboratory, where
they were kept at cold temperature (4 ( 1 °C). All orange juice samples ∑(x
k)1
i,k - xji)2
were first allowed to equilibrate to room temperature (25 °C) before xi,SNV ) (xi,k - xji)/ (1)
vis-NIRS analysis. (m - 1)
Soluble Solids Content and pH in Orange Juice J. Agric. Food Chem., Vol. 54, No. 20, 2006 7439

Figure 1. (a) Original absorbance spectra of orange juices; spectra processed by (b) WPT, (c) SNV, and (d) S. Golay first-der.

Figure 2. Calibration curves for WPT, SNV, and S. Golay first-der in models for SSC (a) and pH (b).

where xi,SNV is the transformed element, xi,k is the original element, xji The first step in PLS is to decompose the matrix, and the model is
is the mean of spectrum i, k ) 1, 2, ..., m, m is the number of variables
in the spectra, i ) 1, 2, ..., n, and n is the number of the validation set. X ) TP + E (2)
Another preprocessing method first derivative is used to remove
background and increase spectral resolution. Savitzky-Golay (23), Y ) UQ + F (3)
which is a moving window averaging method, is used as a smoothing In these equations, T and U are the score matrices of the X matrix and
method in the first derivative, and the number of smoothing is three. It the Y matrix, P and Q are the loading matrices of the X matrix and
is very crucial to select the proper differentiation width (24). Generally, the Y matrix, and E and F are the errors that come from the process of
the width should not exceed 1.5 times the half-width of the absorbance PLS regression.
peak in the spectra. The second step is that T and U are processed by linear regression.
Partial Least-Squares Regression Analysis. Partial least-squares It must build the linear correlation
(PLS) regression with full cross-validation was conducted on the spectra
at 400-1000 nm to analyze the SSC and pH of orange juices. PLS is U ) BT + E (4)
a bilinear modeling method in which the original independent informa-
tion (X variable) is projected onto a small number of latent variables where B is the matrix of diagonally regressed coefficients. To reach
(LVs) to simplify the relationship between X and Y for predicting with this object, the coordinate of T is rotated.
the smallest number of LVs (25). The Y variable is actively used in The cross-validation was performed on the calibration samples based
assessing the LVs to ensure that the first one is most relevant for on excluding a certain number of observations for the calibration model
predicting the Y variable. and used to determine the optimal number of PLS LVs. Due to
7440 J. Agric. Food Chem., Vol. 54, No. 20, 2006 Cen et al.

Figure 3. Reference versus predicted values for (a−c) SSC and (d−f) pH by WPT, SNV, and S. Golay first-der.

overlapping of the overtones from the different groups, the PLS method Table 2. Calibration and Cross-Validation Results for SSC and pH in
was applied to convert the complex spectral data into analytical Orange Juices from the Visible and NIR Spectraa
parameters. Thus, the number of significant PLS LVs was chosen by
using the predicted residual error sum of squares (PRESS) value for SSC (Brix %) pH
every possible LV based on cross-validation (26, 27). The PRESS value range 2.00−13.75 3.30−4.29
was the sum of the squared difference between the predicted and the mean 5.77 3.69
known concentrations. model 1 2 3 4 5 6
SNV, S. Golay first-der, and the whole calibration and validation LVs 4 9 3 7 3 4
calibration
process were achieved in The Unscrambler 9.5 (CAMO Process AS),
slope 0.92 0.98 0.94 0.93 0.93 0.93
a statistical software for multivariate data analysis and experimental offset 0.40 0.13 0.29 0.25 0.26 0.26
design. correlation 0.96 0.99 0.97 0.97 0.96 0.96
SEC 0.97 0.55 0.84 0.06 0.06 0.06
RESULTS AND DISCUSSION RMSEC 0.97 0.54 0.83 0.06 0.06 0.06
bias (10-7) 2.2 2.03 2.41 −0.36 −0.82 −1.05
Features of Visible and NIR Spectra. Figure 1a shows the cross-validation
25 original spectra for 25 samples selected randomly from slope 0.93 0.96 0.94 0.93 0.92 0.93
offset 0.41 0.19 0.36 0.25 0.28 0.26
orange juice samples. Considerable noise appeared in the
correlation 0.95 0.97 0.96 0.95 0.96 0.93
wavelength region at 400-500 and 950-1000 nm, and it could SECV 1.16 0.83 1.03 0.07 0.06 0.08
be eliminated by different preprocessing methods. The shape RMSECV 1.15 0.82 1.03 0.07 0.06 0.08
of the spectra was quite homogeneous, and no outliers were bias 0.02 −0.01 0.01 −0.0008 −0.0001 −0.0018
identified by the naked eye. Some peaks and valleys representing
a Models: WPT- 1 and -4, SNV-2 and -5, S. Golay first-der-3 and -6.
the characteristics of orange juices including hidden information
of different elements and their quantities were obviously shown The light does not always travel the same distance in the sample
in the spectra. before it is detected. A longer light traveling path corresponds
There are consistent baseline shifts and bias in the spectra to a lower relative reflectance value, because more light is
due to the light scattering or concentration variation of samples. absorbed. This causes a parallel shift of the spectra. This kind
Soluble Solids Content and pH in Orange Juice J. Agric. Food Chem., Vol. 54, No. 20, 2006 7441

Figure 4. Regression coefficients obtained after optimal number of LVs in PLS models. The number of PLS LVs (a−c) was 4, 9, and 3 for SSC and
(d−f) 7, 3, and 4 for pH preprocessed by WPT, SNV, and S. Golay first-der, respectively.

of variation is not useful for the calibration models and needs Multivariate Calibration. The calibration models for SSC
to be eliminated by data preprocessing techniques. and pH with preprocessed spectra were developed using PLS
Panels b, c, and d of Figure 1 show the spectra of orange regression with cross-validation. One hundred and four samples
juices processed by WPT, SNV, and S. Golay first-der. It is were split into two groups. Seventy-three samples were used
quite obvious that the peak and valley positions of the processed for the calibration, and the residual 31 were used for the
spectra are corresponding to the valleys or peaks in the original validation. It is eligible to build and evaluate the models. To
spectra (Figure 1a). It can be seen that denoising of WPT did evaluate the results, standard error of calibration (SEC) and
not eliminate any important feature of the spectra, and then all correlation coefficient (rc) and standard error of cross-validation
of the relevant chemical information was retained for modeling. (SECV) and correlation coefficient (rcv) were considered.
The SNV method eliminated the baseline shifts of the spectra Good models should have lower SEC and SECV and higher
(22) and made the peaks and valleys more obvious, such as the correlation coefficients (r) but smaller differences between SEC
wavelength range of 600-700 nm. S. Golay first-der caused and SECV, because large ones indicate that too many LVs are
great changes in the slopes of the raw spectra, and many introduced in the model and the noises are also modeled. In
overlapped peaks could be differentiated. addition, the root-mean-square error of cross-validation
The main constituents in orange juices are organic acids (RMSECV) was used to determine the optimal model without
(citric, tartaric, malic acid, etc.), sugars (fructose, sucrose, etc.), “overfittedness” or “underfittedness” (27).
vitamins, and some natural pigments (flavonoids and caro- In Figure 2, the RMSECV curves for the mean-centered,
tenoids). In the visible region, pigments would produce some WPT-, SNV-, and S. Golay first-der-corrected samples display
influences on the variation of absorbance in the spectra. small values (close to 0) with as few LVs as possible at 4, 9,
However, the relationship between pH and some pigments can and 3 LVs for SSC and 7, 3, and 4 LVs for pH, respectively.
provide information to build PLS calibration about the variation The corresponding vectors were then used for the secondary
of pH value in the visible region. Also, the chemical structure samples that had been corrected and mean-centered on the basis
of sugars and some organic acids was related with the NIR of the mean vector for the corrected samples. From Figure 2,
region. RMSECV does not appear to change significantly after the
7442 J. Agric. Food Chem., Vol. 54, No. 20, 2006 Cen et al.

certain number of LVs, which were used as PLS LVs, absorbance and the appearance of some peaks such as wave-
simplifying calibration models. length at 520 and 620 nm may be affected by carotenoids and
Reasonable models were obtained by PLS multivariate flavonoids in orange juices. However, when models are built,
calibration with full cross-validation, and the results are shown the whole wavelength is used in PLS analysis, which considered
in Table 2. This calibration revealed that some differences the concentration matrix (Y variable). Thus, SSC and pH were
existed among different models with different preprocessing also related to these peaks in the visible region. The band at
methods. The models for SSC and pH using SNV yield the best 930 nm possibly resulted from C-H stretch third overtone from
results in the cross-validation. Whether for SSC or pH, the sugars, and spectral variation at 990 nm was produced by the
SECV, RMSECV, and bias are lowest, and correlation coef- O-H stretch second overtone from sugars, organic acids, and
ficients are highest in models built by SNV-processed spectra. flavonoids (29). The attempt to find a fingerprint for SSC and
It is interesting to note that the number of LVs selected by PLS pH can be valuable due to the lack of this information in visible
regression models varied. This effect is probably due to the and NIR spectroscopy of orange juice or other materials.
differences in the signal-to-noise ratio on preprocessing proce- It is concluded that PLS regression as chemometrics combined
dures. with the vis-NIRS technique for evaluation of chemical
Prediction Results of PLS Regression Analysis for SSC compositions in orange juices is successfully applied by simple
and pH. On the basis of the above, three PLS models were models and feature extraction techniques. The statistic param-
applied to predict 31 residual samples after similar mathematical eters of calibration and validation showed that the PLS
preprocessing to the calibration ones. The prediction results of regression model was an available alternative for quality
SSC and pH in orange juices using three different preprocessing detection of orange juices based on vis-NIRS. Multiplicative
methods are shown in Figure 3. The regression line represented signal correction was proposed and used as an effective
ideal results, and so the closer the points are to this, the better preprocessing method in that the calibration models for the
was the model. The high correlation coefficients (r) and low internal quality of orange juices were optimized. It is also
standard error of prediction (SEP) and the root-mean-square necessary to select an optimal preprocessing method to improve
error of prediction (RMSEP) indicated that these models built the accuracy and reliability of regression models. On the basis
by PLS with WPT, SNV, and S. Golay first-der could qualify of PLS regression coefficients, the fingerprint representing
for predicting SSC and pH of orange juices. features of orange juice or reflecting sensitivity to some elements
All of the models developed had r in prediction >0.93, and at a certain band was proposed. The fingerprint analysis is very
the r of 0.98 and 0.96 in PLS with the SNV model for SSC useful in the field of food chemistry, and further research on
and pH is the highest. SEP and RMSEP were also calculated, other materials is needed to improve the reliability and precision
showing the evaluating standard for the quality of models and of this technology.
the predictive abilities. In the case of PLS regression analysis
with WPT and S. Golay first-der, the SEP values between LITERATURE CITED
measured and predicted SSC were 1.09 and 1.02, respectively
(1) Luis, E. R.; Fredrick, S. F.; Michael, A. M.; Elizabeth, M. C.
(see Figure 3b). However, in the case of SNV preprocessing, Rapid analysis of sugars in fruit juices by FT-NIR spectroscopy.
the SEP was improved to 0.68, which is much lower than that Carbohydr. Res. 2001, 336, 63-74.
in the two others, as well as RMSEP. Interestingly, the (2) Saavedra, L.; Garcia, A.; Barbas, C. Development and validation
application of the SNV preprocessing method for pH prediction of a capillary electrophoresis method for direct measurement of
was also the best in multivariate statistical analysis. The 0.06 isocitric, citric, tartaric and malic acids as adulteration markers
SEP and 0.06 RMSEP were the lowest (see Figure 3e). In view in orange juice. J. Chromatogr. A 2000, 881, 395-401.
of these prediction results, it is very important to select a prior (3) Fish, W. W.; Perkins-Veazie, P.; Collins, J. K. A quantitative
preprocessing method for the analysis of spectral data. In assay for lycopene that utilizes reduced volumes of organic
addition, it is possible to find a good calibration equation solvents. J. Food Compos. Anal. 2002, 15, 309-317.
(4) Breithaupt, D. E. Simultaneous HPLC determination of caro-
concerned with SSC and pH in orange juice based on the vis-
tenoids used as food coloring additives: applicability of acceler-
NIRS technique with proper chemometrics by employing a much ated solvent extraction. Food Chem. 2004, 86, 449-456.
more explanatory variable. (5) Hu, Y. X.; Yuan, S. X.; Li, J. N.; Zhang, P.; Zhu, Q. M. Detection
Fingerprint Analysis Based on Regression Coefficients. for the adulteration of citrus juice by fluorescence spectra. Sci.
With different preprocessing techniques, the fingerprints reflect- Technol. Food Ind. 2005, 26, 166-168.
ing the characteristics of spectra for SSC and pH were obtained (6) He, Z.; Gao, H.; Yuan, L. J.; Lu, S. F.; Meng, H.; Li, X. Y.;
on the basis of regression coefficients (Figure 4). It could be Zeng, Y. Pulse injection analysis with chemiluminescence
seen that the shape and trend of fingerprints were similar for detection: determination of citric acid using tris-(2,2-bipyridine)
ruthenium(II). Talanta 1998, 47, 301-304.
SSC or pH, respectively, although different signal correction
(7) Scordino, M.; Di Mauro, A.; Passerini, A.; Maccarone, E.
methods were used before calibration models were built. From Phloroglucinol from phlorin hydrolysis for testing quality of
Figure 4a-c, it was found that wavelengths of 410, 520, 640, commercial orange juices and beverages. J. Agric. Food Chem.
930, and 990 nm might be of particular importance for the SSC 2004, 52, 7469-7474.
calibration. Furthermore, the regression coefficients shown in (8) Fuzzati, N. Analysis methods of ginsenosides. J. Chromatogr.
the figures calculated from different numbers of LVs give B 2004, 812, 119-133.
significant weights at the wavelength regions referred to above. (9) Cozzolino, D.; Murray, I. Identification of animal meat muscles
The regression coefficients shown in Figure 4d-f also have by visible and near infrared reflectance spectroscopy. Lebensm.-
strong peaks and valleys at certain wavelengths, such as 410, Wiss. -Technol. 2004, 37, 447-452.
490, 520, 620, 640, and 990 nm, related to pH. (10) He, Y.; Li, X. L.; Shao, Y. N. Quantitative analysis of the brands
of apple using near infrared spectroscopy by principal component
In the visible region, the pigments in orange juice have some analysis and BP model. AI2005: AdVances in Artificial Intel-
influences, and changes in the pH of the juice could produce ligence; Lecture Notes in Artificial Intelligence; Zhang, S. C.,
shifts in the absorbance of particular wavelengths in the visible Jarvis, R., Eds.; Springer-Verlag: Berlin, Germany, 2005; pp
region related with those pigments (28). Thus, the variation of 1053-1056.
Soluble Solids Content and pH in Orange Juice J. Agric. Food Chem., Vol. 54, No. 20, 2006 7443

(11) Tsuchikawa, S.; Hamada, T. Application of time-of-flight near (22) Chu, X. L.; Yuan, H. F.; Lu, W. Z. Progress and application of
infrared spectroscopy for detecting sugar and acid contents in spectral data pretreatment and wavelength selection methods in
apples. J. Agric. Food Chem. 2004, 52, 2434-2439. NIR analytical technique (in Chinese). Prog. Chem. 2004, 16,
(12) Pedro, A. M. K.; Ferreira, M. M. C. Nondestructive determination 528-542.
of solids and carotenoids in tomato products by near-infrared (23) Liang, Y. C.; Yi, Z. S. The Handbook of Analytical Chemistry:
spectroscopy and multivariate calibration. Anal. Chem. 2005, 77, Chemistry Metrology; Chemical Industry Press: Beijing, China,
2505-2511. 2001.
(13) Pintes, M. J. C.; Santos, S. R. B.; Araujo, M. C. U.; Almeida, (24) Xu, G. T.; Yuan, H. F.; Lu, W. Z. Development of modern near
L. F.; Lima, R. A. C.; Gaiao, E. N.; Souto, U. T. C. P. infrared spectroscopic techniques and its applications. Spectrosc.
Classification of distilled alcoholic beverages and verification Spectral Anal. (Beijing, China) 2000, 20, 134-142.
of adulteration by near infrared spectrometry. Food Res. Int. (25) Xing, J.; Linden, V. V.; Vanzeebroeck, M.; Baerdemaeker, J.
2006, 39, 182-189. D. Bruise detection on Jonagold apples by visible and near-
(14) Cozzolino, D.; Murray, I.; Chree, A.; Scaife, J. R. Multivariate infrared spectroscopy. Food Control 2004, 16, 357-361.
determination of free fatty acids and moisture in fish oils by (26) Szlyk, E.; Szydlowska-Czerniak, A.; Kowalczyk-Marzec, A. NIR
partial least-squares regression and near-infrared spectroscopy. spectroscopy and partial least-squares regression for determina-
Lebensm.-Wiss. -Technol. 2005, 38, 821-828. tion of natural R-tocopherol in vegetable oils. J. Agric. Food
(15) Rambla, F. J.; Garrigues, S.; Guardia, M. D. L. PLS-NIR Chem. 2005, 53, 6980-6987.
determination of total sugar, glucose, fructose and sucrose in (27) Naes, T.; Isaksson, T.; Fearn, T.; Davies, A. M. A User-friendly
aqueous solutions of fruit juices. Anal. Chim. Acta 1997, 344, Guide to MultiVariate Calibration and Classification; NIR
41-53. Publications: Chichester, U.K., 2002.
(16) Dou, Y.; Mi, H.; Zhao, L.; Ren, Y.; Ren, Y. Determination of (28) Laleh, G. H.; Frydoonfar, H.; Heidary, R.; Jameei, R.; Zare, S.
compound aminopyrine phenacetin tablets by using artificial The effect of light, temperature, pH and species on stability of
neural networks combined with principal components analysis. anthocyanin pigments in four Berberis species. Pakistan J. Nutr.
Anal. Biochem. 2006, 351, 174-180. 2006, 5, 90-92.
(17) Walczak, B.; van den Bogaert, B.; Massart, D. L. Application (29) Sasic, S.; Ozaki, Y. Short-wave near-infrared spectrosocopy of
of wavelet packet transform in pattern recognition of near IR biological fluids. 1. Quantitative analysis of fat, protein, and
data. Anal. Chem. 1996, 68, 1742-1747. lactose in raw milk by partial least-squares regression and band
(18) Sorensen, L. K.; Jepsen, R. Assessment of sensory properties of assignment. Anal. Chem. 2001, 73, 64-71.
cheese by near-infrared spectroscopy. Int. Dairy J. 1998, 8, 863-
871. Received for review June 16, 2006. Revised manuscript received July
(19) Drai, R.; Khelil, M.; Benchaala, A. Time frequency and wavelet 28, 2006. Accepted August 1, 2006. This study was supported by the
transform applied to selected problems in ultrasonics NDE. Teaching and Research Award Program for Outstanding Young
NDT&E Int. 2002, 35, 567-572. Teachers in Higher Education Institutions of MOE, P.R.C., Natural
(20) Chen, J.; Shi, Y.; Shi, S. Noise analysis of digital ultrasonic Science Foundation of China (Project 30270773), Specialized Research
nondestructive evaluation system. Int. J. Pressure Vessels Piping Fund for the Doctoral Program of Higher Education (Project
1999, 76, 619-630. 20040335034), and Science and Technology Department of Zhejiang
(21) FECIT Sci-Tech. The Theory of WaVelet Analysis and MATLAB Province (Projects 2005C21094 and 2005C12029).
7 Application; Publishing House of Electronics Industry: Beijing,
China, 2005; 238 pp. JF061689F