LWT - Food Science and Technology 66 (2016) 86e92

Contents lists available at ScienceDirect

LWT - Food Science and Technology

journal homepage: www.elsevier.com/locate/lwt

Monitoring of alcohol strength and titratable acidity of apple wine

during fermentation using near-infrared spectroscopy
Bangzhu Peng a, *, Na Ge a, Lu Cui b, Hui Zhao b
a
Key Laboratory of Environment Correlative Dietology, Ministry of Education, College of Food Science & Technology, Huazhong Agricultural University,
430070, Wuhan, Hubei, China
b
College of Food Science & Engineering, Northwest A & F University, 712100, Yangling, Shaanxi, China

a r t i c l e i n f o a b s t r a c t

Article history: Rapid development in apple wine industry has raised a growing need for effective fermentation process
Received 25 May 2015 monitoring. The alcohol strength and titratable acidity have been used as indicators for process control
Received in revised form and optimization in apple wine making. The objective of this study was to investigate the efcacy of
29 September 2015
using near-infrared (NIR) spectroscopy to monitor and assess the process parameters during apple wine
Accepted 5 October 2015
Available online 9 October 2015
fermentation. A simple and rapid NIR spectroscopy method was developed for simultaneous determi-
nation of alcohol strength and titratable acidity in apple wine. The selected spectral regions of 6101.9
e5446.2 cm
1, and 11,995.4e7498.1 cm
1 were pretreated by second derivative (SD) and straight line
Keywords:
Near-infrared spectroscopy
subtraction (SLS) for alcohol strength and titratable acidity, respectively, prior to developing calibration
Apple wine models using partial least squares (PLS) regression with cross-validation. The highest R2c and the lowest
Monitoring RMSECV in the calibration set were obtained for the alcohol strength (0.923 and 4.63 mL/L) and for the
Fermentation titratable acidity (0.930 and 0.264 g/L). The NIR calibration models showed good correlation of deter-
Process control mination and low predictive errors. Application of the NIR calibration models demonstrated the feasi-
bility of NIR spectroscopy to be used as a quality control tool for monitoring the apple wine fermentation
process.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
Apple wine is widely used as a fermented alcoholic beverage in
most countries. The steadily-growing diversity of commercial apple
wine, coupled with increasing consumer demand, has prompted a
need to monitor their major chemical composition changes for
process control and quality assessment in order to better manage
fermentation process. Alcohol strength and titratable acidity have
been used as indicators during apple wine fermentation process for
quality control and optimization in apple wine industry; hence, it is
of great importance to monitor the changes of alcohol strength and
titratable acidity in fermentation process for improving apple wine
quality. A variety of methods have been developed for monitoring
the parameters of alcohol strength and titratable acidity in previous
research, including chemical assays, pH meters, and high pressure
liquid chromatography, etc. (S.A.C., 2008; Satora, Tarko, Sroka, &
Blaszczyk, 2014; Wilson, Maguer, Duitschaever, Buteau, & Allen,
* Corresponding author.
E-mail address: bzpeng@mail.hzau.edu.cn (B. Peng).
2003; Ye, Yue, Yuan, & Li, 2014; Zhang, Fang, & Li, 2011), but
these methods generally had the disadvantages of sample prepa-
ration, reagent consumption and complicated operation with time-
consuming that might hamper the quality of nal apple wine. With
the development of brewing technology and the demand of quality
control, it is important to perform fast and accurate determination
of target composition for cost saving and online monitoring prod-
uct quality (Blanco, Peinado, & Mas, 2004).
Nowadays, apple wine brewing industry needs to introduce and
develop more advanced analysis method online in order to get
reliable feedback data in time, which applied to guide the
fermentation process. The monitoring of a fermentation process
often requires fast and frequent measurements, which makes Near-
infrared (NIR) spectroscopy an excellent candidate for online
analysis to provide rapid determination and straightforward
monitoring. Bioprocess monitoring is becoming one of the most
challenging applications of NIR spectroscopy (Lachenmeier, 2007).
With the application of chemometric techniques, a single spectrum
can be subjected to many different calibration models, to measure
many compositions in various types of food (Shen et al., 2010). NIR
spectroscopy is a widely used rapid and non-destructive technique

http://dx.doi.org/10.1016/j.lwt.2015.10.018
0023-6438/ 2015 Elsevier Ltd. All rights reserved.
 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92
for quality measurements and control in food and agricultural
product analyses (Antonucci et al., 2011; Chen, Ding, Cai, Sun, &
Zhao, 2012; Chen, Zhao, Fang, & Wang, 2007; Grassi, Amigo,
Lyndgaard, Foschino, & Casiraghi, 2014; Pouliot, Paquin, Martel,
Gauthier, & Pouliot, 1997; Xie, Ye, Liu, & Ying, 2011), and is
currently an ideal alternative to traditional analytical techniques for
monitoring bioprocess (McLeod et al., 2009). With the application
of chemometric techniques, a single spectrum can be subjected to
many different calibration models, to measure many compositions
(McLeod et al., 2009), and this modern analytical technology has
been used as a well-suited method for general process real-time
monitoring, which is of great interest for many wine makers to
conduct process control. Di Egidio, Sinelli, Giovanelli, Moles, and
Casiraghi (2010) obtained good calibration models for the predic-
tion of the main compositional changes with FT-NIR to evaluate
alcoholic fermentation online in red wine. Buratti et al. (2011)
demonstrated that the non-destructive method of NIR are suit-
able for the monitoring of must-wine fermentation giving crucial
information about the quality of the nal product in agreement
with chemical parameters. Giovenzana, Beghi, and Guidetti (2014)
carried out a preliminary study to verify the possibility of
employing a device based on NIR spectroscopy, directly on the
production line of craft beer. Grassi et al. (2014) investigated the
capability of FT-NIR spectroscopy to monitor and assess process
parameters in beer fermentation with two different yeast strains at
different temperatures, and the results showed that FT-NIR spec-
troscopy demonstrated to be a perfectly suitable quantitative
method to be implemented in the production of beer. NIR spec-
troscopy has already shown promise as a rapid and non-destructive
method for determining various compounds simultaneously in the
fermentation industry (Liang et al., 2013). Nevertheless, less
research exists on the application of NIR spectroscopy for moni-
toring and assessment of changes in relevant physico-chemical
parameters in the apple wine fermentation process. Therefore,
much more intensive work is needed to really evaluate the feasi-
bility of NIR spectroscopy in monitoring apple wine fermentation
process. The objectives of this study were to develop fourier
transform near infrared (FT-NIR) spectroscopy calibration models
for predicting alcohol strength and titratable acidity of apple wine
in order to monitor fermentation process. In addition, the feasibility
of using FT-NIR spectroscopy to monitor these two indicators
simultaneously during the fermentation of apple wine was
explored.
2. Materials and methods
2.1. Apple wine fermentation and sampling
Ripe Fuji apples used in this study were harvested from different
commercial orchards located at Luochuan county in northwest
China, and stored in a cold room at 4  C until use. Apple juice was
obtained by squeezing the mixed apples from different orchards
from LZ juice extractor with a vacuum pressing and ltration sys-
tem in room temperature (Gongda company, Wenzhou, China),
during which the sodium bisulte of 100 mg/L was added to apple
juice in order to inhibit the bacteria growth immediately (Peng,
Yue, & Yuan, 2008). The total sugar and pH of all juice samples
for fermentation were then adjusted to 200 g/L with saccharose and
3.4 with 0.5 M hydrochloric acid, respectively. A total of nine apple
wine fermentation trials were conducted using three different
yeasts (Saccharomyces cerevisiae 1750, PF12 and WP45) at three
temperatures of 17, 20 and 23  C according to dry type brewing
technology in laboratory. The apple wine samples were collected
right after yeast inoculation (0 h, starting time), and then every 12 h
until the 10th day of fermentation in each trial. When sampling,
87
about 30 mL of samples were discharged out directly from fer-
mented supernatant. One hundred and sixty samples were
collected from eight trials for developing the models of quantifying
the alcoholic strength and titratable acidity, while another trial
with twenty collected samples was used to evaluate the feasibility
of FT-NIR spectroscopy in monitoring apple wine fermentation
process.
2.2. Chemical analysis
The performance of a quantitative NIR model is mainly depen-
dent on the performance of reference method (Bock & Connelly,
2008). Poor precision and accuracy of the reference method will
limit the performance of the NIR model. Therefore, it is important
to ensure the targeted parameter as close to the true value as
possible. In this study, samples were centrifuged at 5000 rpm for
5 min and supernatants were collected for chemical analysis.
Alcoholic strength was measured by using an PT-1 alcoholmeter
(Pute company, Beijing, China), immersed in a distillate at a tem-
perature of 20  C; titratable acidity was determined by acidebase
titration method according to SAC standard GB/T 15038 (S.A.C.,
2008) and the results were expressed as malic acid. All reagents
used were of analytical grade, and related solutions were prepared
by ultrapure water produced with a millipore purication system
(Millipore, Mississauga, Canada) in the experiments. All analyses
were done in triplicate.
2.3. Spectrum collection
FT-NIR spectra collected directly from the supernatant of apple
wine samples using a MPA FT-NIR spectrometer (Bruker Optics,
Germany) in the range of 12,000 to 4000 cm
1, which equipped
with an InGaAs detector and a liquid ber optic probe with a 2 mm
path length. An air background was employed as a probe reference,
and the liquid probe directly immersed into samples for collecting
the spectra at 25  C. The spectral resolution was set at 8 cm
1 with
an interval of 2 cm
1 and the spectra were recorded performing 20
scans for both the reference and tested samples. The spectra were
recorded in triplicate and an average spectrum was collected to
ensure an adequate signal-to-noise ratio for the subsequent anal-
ysis. The NIR instrument control, spectra scanning, and initial data
processing were performed by using OPUS software version 5.5
(Bruker Optics, Germany).
2.4. Preprocessing of spectra and chemometric analysis
NIR spectra preprocessing in multivariate calibration is still trial
and error due to the lack of sufcient prior information and an
incomplete knowledge of the raw data (Xu et al., 2008). How to
select a proper preprocessing method depends largely on the
expertise and experience of practitioners. To establish best cali-
bration model, spectra data from samples need to conduct different
treatments for reducing noise interference and increasing the res-
olution of small absorbance bands (Huang, Yu, Xu, & Ying, 2008).
Chemometric analysis is usually used as a tool to correlate the
effective information in the NIR spectra signal to a number of latent
variables or factors, constrained by a set of calibration reference
data (Massart et al., 1997). Partial least squares (PLS) regression is a
well-established chemometric technique whereby factors are
derived by taking into account the variation in the spectral data
that is relevant for explaining variation in the characteristics of
interest in the original samples (Williams, 2001).
In this study, spectra preprocessing was made by straight line
subtraction (SLS), rst derivative (FD), second derivative (SD),
vector normalization (VN), minemax normalization (MMN),
88 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92

Table 1
Analysis for chemical reference values of alcohol strength and titratable acidity in samples.

Parameter Set n Range Mean CV (%) PSE

Alcohol strength (mL/L) Calibration 120 0e126.3 6.63 67.6 1.92

Prediction 40 0e117.5 6.13 58.7 1.63
Titratable acidity (g/L) Calibration 120 2.08e8.92 5.35 32.7 0.161
Prediction 40 3.07e6.80 4.60 24.3 0.172

n: the number of samples; CV: coefcient of variation; PSE: pooled standard error.

multiplicative scattering correction (MSC), and the combination of 3. Results and discussion
every two methods using OPUS software (v.5.5 Bruker Optics,
Germany). The number of latent factors used in the calibration 3.1. Chemical analysis
models is of critical importance to avoid under or over tting of the
data and to ensure that the models are optimized for the intended Samples for establishing NIRS model should be representative of
purpose of the procedure (Inon, Llario, Garrigues, & de la Guardia, the commercial fermentation process of apple wine, and have the
2005; Valderrama, Braga, & Poppi, 2007). expected variability within the scope of the NIR calibration proce-
dure. The calibration and prediction sets contained 120 and 40
samples, respectively, which were selected by the KennardeStone
2.5. Establishment and evaluation of NIRS models algorithm in each set that covered the whole range of each chem-
ical parameter (Kennard & Stone, 1969; Valderrama et al., 2007).
For establishment of NIRS model, calibration is generally per- The samples of prediction set were selected as external validation
formed by developing a calibration model that relates target (not used in calibration procedures) to evaluate the prediction
composition concentration to absorbance spectra for a set of ability of calibration models.
reference samples in calibration set (Cozzolino, Parker, Dambergs, The range, mean, coefcient of variation (CV) and pooled stan-
Herderich, & Gishen, 2006). In this study, the calibration models dard error (PSE) for the alcohol strength and titratable acidity
of alcohol strength and titratable acidity were developed using PLS measured in both calibration and prediction sets were listed in
regressions with leave-one-out cross validation, which it makes a Table 1. A calibration set containing 120 samples was used for
subset of the samples leaving one out and trying to predict the one model development in the ranges from 0 to 126.3 mL/L for alcohol
that was left out (Norgaard et al., 2000). The performance of cali- strength, and from 2.08 to 8.92 g/L for titratable acidity, respec-
bration models was evaluated in terms of determination coefcient tively. The ranges in values measured were in accordance with
(R2) and root mean square error of cross-validation (RMSECV), typical values expected in apple wines (Satora et al., 2014; Ye et al.,
among which the R2 is a way of describing the strength of the 2014). Concentration spots of alcohol strength and titratable acidity
relationship between two variables, and the larger the R2 is, the in samples were plotted in Fig. 1. As can be seen from the Fig. 1,
more variability is explained by the calibration model; the RMSECV there is a positive relationship between alcohol strength and
is usually dened as the standard deviation of differences between titratable acidity. This is because both of the two substances are
spectral data and reference values in the cross-validation sample yeast metabolites in fermentation process, and can thus accumu-
set (Di Egidio, Olivieri, Woodcock, & Downey, 2011; Williams, late continuously in apple wine samples. The statistics data showed
2001). Generally, a better model should have higher R2 and a considerable degree of variations in the content of alcohol
smaller RMSECV. In addition, to evaluate the predictive capability of strength and titratable acidity, as indicated by the range and CV
the calibration model, root mean square error of prediction
(RMSEP), which indicates the total error in prediction, should be
acceptable when the model is applied to generate predictions
outside the calibration set. Moreover, the ratio of standard devia-
tion to the standard error of prediction (RPD) was also employed as
an indicator of NIRS predictions in this study. And higher RPD
values usually indicate better NIR predictions (Williams, 2001).
Generally, a good model should possess higher values of R2 and
RPD, and lower values of RMSEC and RMSEP (Bao et al., 2014). Only
optimal models are selected and discussed in this study.
The related equations dened as
0 1
B Pn
B yTi
yPi 2 C
C
R2 B1
Pi1 C;
@ n
yTi
ya 2 A
i1

s s
Pn 2 Pm 2
i1 yTi
yPi i1 yTi
yPi
RMSECV ; RMSEP
n m

where yTi is the reference value of the parameter for the i th sample
of the calibration set or prediction set; yPi is the predicted value
obtained by NIRS model; ya is the average value of reference; n, m
are the number of samples in calibration set and prediction set,
respectively.
Fig. 1. Concentration spots of alcohol strength and titratable acidity in samples for
developing FT-NIRS models of alcohol strength and titratable acidity in order to
monitor apple wine fermentation process.
 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92 89

Table 2
Statistics of calibration models of alcohol strength and titratable acidity in apple wine samples.

Parameter Spectral region Preprocessing method Rank Calibration set Prediction set

n RMSECV R2c n RMSEP R2p RPD

1
Alcohol strength (mL/L) 6101.9e5446.2 cm SD 6 115 4.63 0.923 40 4.25 0.993 7.12
Titratable acidity (g/L) 11,995.4e7498.1 cm
1 SLS 6 114 0.264 0.931 40 0.213 0.973 6.86

SD: second derivative; SLS: straight line subtraction; RMSECV: root mean square error of cross validation; R2c : determination coefcient for calibration set; R2p : determination
coefcient for prediction set; RPD: residual predictive deviation.

values in Table 1. The large CV values of alcohol strength (67.6%,
58.7%, respectively for calibration and prediction sets) and titrat-
able acidity (32.7%, 24.3%, respectively for calibration and predic-
tion sets) were possibly related to different fermentation
conditions being studied. Therefore, the variability of sample data
in both calibration and prediction sets was considered suitable for
developing stable and reliable NIR calibrations for alcohol strength
and titratable acidity.
3.2. Selection and preprocessing of NIR spectra
Preprocessing of raw near-infrared spectra (NIRS) is indispens-
able because the measured spectra are subject to signicant noises,
baselines and other irrelevant information, which may affect the
predictive ability of NIRS model (Blanco et al., 2004). Hence, it is
imperative to perform the spectral range selection and pretreat-
ment for visualizing and extracting relevant information from the
spectra best suited for calibration model. In this study, the auto-
mation optimization function of the OPUS-QUANT2 software was
used to optimize the calibration conditions for developing best
calibration models of alcohol strength and titratable acidity. The
best conditions including the spectral region, preprocessing
method and rank, were selected based on the smaller rank value
and RMSECV for the calibration models according to PLS algorithm.
The statistics of optimal conditions are given in Table 2. The best
spectral regions selected for developing calibration models of
alcohol strength and titratable acidity were 6101.9e5446.2 cm
1,
11,995.4e7498.1 cm
1, respectively. The proles of selected spectral
regions preprocessed by second derivative (SD) and straight line
subtraction (SLS) for alcohol strength and titratable acidity,
respectively, are shown in Figs. 2 and 3. From Figs. 2 and 3, more
positive information and less noise were identied and extracted
from the differences of the spectra pretreated by SD or SLS, and thus
these pretreatments had contributed most to precision of each
model.
3.3. Calibration and validation of models
The calibration models of alcohol strength and titratable acidity
were developed using PLS regressions with leave-one-out cross
validation. The PLS regression is a technique that constructs new
predictor factors derived by taking into account the variation in

Fig. 2. The prole of selected spectral regions of 6101.9 ~ 5446.2 cm
1 before and after pretreatment by second derivative (SD) for developing calibration model of alcohol strength
in apple wine fermentation process (a: the original selected spectra; b: the spectra after pretreatment by SD).
90 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92

Fig. 3. The prole of selected spectral regions of 11995.4~7498.1 cm
1 before and after pretreatment by straight line subtraction (SLS) for developing calibration model of titratable
acidity in apple wine fermentation process (a: the original selected spectra; b: the spectra after pretreatment by SLS).

spectral data that is relevant for explaining variation in the char-
acteristics of interest in original samples (Li, Zhang, & Atungulu,
2014; Williams, 2001), and it is also one of the standard methods
used to develop calibration equations to predict chemical compo-
sitions of a sample from NIR spectra (Reeves, 2001).
Statistics for the calibration models developed for alcohol
strength and titratable acidity in apple wines using FT-NIR spectra
are shown in Table 2. Outliers in the calibration set were identied
and excluded based on extreme leverages (Lillhonga & Geladi,
2005). The lowest RMSECV and the highest R2c in the calibration
set were obtained for alcohol strength (4.63 mL/L and 0.923) with
removal of 5 outliers, and titratable acidity (0.264 mg/L and 0.930)
with removal of 6 outliers. Generally, values of R2 greater than 0.9
indicate excellent quantitative information for calibration model
(Shenk & Westerhaus, 1996). To improve the stability of the models,
the rank must be appropriately chosen to avoid overt during
calibration. Fig. 4 shows the correlation between RMSECV and Rank
in calibration by PLS regression with leave-one-out validation, and

Fig. 4. Correlation between RMSECV and Rank in calibration set by PLS regression with leave-one-out cross validation for developing FT-NIRS models (a: for alcohol strength; b: for
titratable acidity).
 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92 91

Fig. 5. Correlation statistics between chemical reference value and FT-NIRS prediction value in calibration and prediction sets. The FT-NIRS models were developed by PLS
regression with leave-one-out cross validation using selected spectral regions of 11,995.4 ~ 7498.1 cm
1 pretreated by SLS, and 6101.9 ~ 5446.2 cm
1 pretreated by SD for alcohol
strength (a) and titratable acidity (b), respectively.

the optimum ranks are both 6 for alcohol strength and titratable suggest that the calibration models have good predictive capacity
acidity calibration models. The correlation plots between chemical for alcohol strength and titratable acidity during apple wine
reference value and FT-NIR prediction value for the two parameters fermentation, and this rapid NIR method could be taken to monitor
of apple wine in calibration set are shown in Fig. 5. The results show the apple wine fermentation process with sufcient feasibility.
that the NIRS calibration models have excellent precision. There-
fore, these two physicochemical parameters in apple wine might be
measured by FT-NIR spectroscopy. 4. Conclusion
In order to further evaluate the prediction capability of the
models, an external validation was conducted by applying FT-NIRS This study developed a simple and rapid NIRS method for
models to the 40 samples of prediction set. Table 2 shows that simultaneous determination of alcohol strength and titratable
adequate small RMSEP values were obtained for alcohol strength acidity content in apple wine. The NIRS models showed a good
and titratable acidity (4.25 mL/L and 0.213 mg/L), meanwhile the R2p agreement between calibration and predication data with adequate
values (0.993 and 0.973) are closed to 1. The rate of prediction to high correlation of determination of R2 and relatively low predic-
deviation (RPD) could be an indicator of model validation and a tive errors of RMSEP. The results demonstrated that NIR spectros-
higher RPD would be deemed better for prediction capability of copy can used as an alternative method to rapidly analyze alcohol
model. A good calibration model should have a RPD value higher strength and titratable acidity, with a very high applicability for
than 3.0 (Williams, 2001). A RPD value of 6.5e8.0 is generally monitoring apple wine fermentation process in order to improve
considered as good for quality control purpose (Botelho, Mendes, & the quality of apple wine.
Sena, 2013). As shown in Table 2, RPD for the calibration models of
alcohol strength and titratable acidity (7.12 and 6.86) were
considered satisfactory. Additionally, the correlation plots between
chemical reference value and FT-NIR prediction value for the two
parameters of apple wine in prediction set are shown in Fig. 5. The
results indicate that the calibration models are adequate as a tool to
determine the alcohol strength and titratable acidity with accurate
performance during apple wine fermentation. Similar results were
obtained by Ye et al. (2014) for the determination of total acidity
with the determination coefcients of 0.98 and the RMSEP of 0.021
in apple wine by FT-NIRS transmission mode.

3.4. Application of models

In order to check the feasibility of NIRS method developed in

this study, one apple wine fermentation trail was monitored by the
calibration models. The alcohol strength and titratable acidity with
time during the apple wine fermentation were predicted using
NIRS calibration and also measured by the chemical reference
methods. The analyses of results are presented in Fig. 6. Fig. 6 shows
that the R2 for the predicted versus measured values of the alcohol
strength and titratable acidity were 0.962 and 0.972, respectively.
The RMSEP of the alcohol strength and titratable acidity were
Fig. 6. Evolution of alcohol strength and titratable acidity in one fermentation run of
3.6 mL/L and 0.22 g/L, respectively. The predicted and measured apple wine at 20  C with S. cerevisiae PF12, the predicted versus measured values of
values of the alcohol strength and titratable acidity showed similar alcohol strength (R2 0.962, RMSEP 3.6 mL/L), and titratable acidity (R2 0.972,
variations over the entire fermentation period. All these results RMSEP 0.22 g/L) using the FT-NIRS models to monitor the fermentation process.
92 B. Peng et al. / LWT - Food Science and Technology 66 (2016) 86e92

Acknowledgments Technometrics, 11, 137e148.

Lachenmeier, D. W. (2007). Rapid quality control of spirit drinks and beer using
multivariate data analysis of Fourier transform infrared spectra. Food Chemistry,
This research was partially supported by the Fundamental 101, 825e832.
Research Funds for the Central Universities (2662015BQ019) and Liang, J., Zhang, D., Guo, X., Xu, Q., Xie, X., Zhang, C. Chen, N. (2013). At-line near-
National Natural Science Foundation of the Republic of China infrared spectroscopy for monitoring concentrations in temperature-triggered
glutamate fermentation. Bioprocess and Biosystems Engineering, 36, 1879e1887.
(20806062, 31301498).The authors acknowledge the anonymous Lillhonga, T., & Geladi, P. (2005). .Replicate analysis and outlier detection in
reviewers very much for their valuable comments and improve- multivariate NIR calibration, illustrated with biofuel analysis. Analytica Chimica
 re
ments. R. Symoneaux, S. Chollet, R. Bauduin, J.M. Le Que , A. Baron. Acta, 544, 177e183.
Li, X., Zhang, A., & Atungulu, G. (2014). Characterization and multivariate analysis of
Impact of apple procyanidins on sensory perception in model cider physical properties of clingstone peaches. Food and Bioprocess Technology, 7,
(part 2): Degree of polymerization and interactions with the matrix 1756e1766.
components. LWT e Food Science and Technology, Volume 57, Issue Massart, D. L., Vandeginste, B. G. M., Buydens, L. M. C., de Jong, S., Lewi, P. L., &
Smeyers-Verbeke, J. (1997). Handbook of chemometrics and qualimetrics: Part B.
1, June 2014, Pages 28e34. Amsterdam: Elsevier.
McLeod, G., Clelland, K., Tapp, H., Kemsley, E. K., Wilson, R. H., & Poulter, G. (2009).
References A comparison of variate pre-selection methods for use in partial least squares
regression: a case study on NIR spectroscopy applied to monitoring beer
fermentation. Journal of Food Engineering, 90, 300e307.
Antonucci, F., Pallottino, F., Paglia, G., Palma, A., Aquino, S., & Menesatti, P. (2011).
Norgaard, L., Saudland, A., Wagner, J., Nielsen, J. P., Munck, L., & Engelsen, S. B.
Non-destructive estimation of mandarin maturity status through portable VIS-
(2000). Interval partial least-squares regression (iPLS): a comparative chemo-
NIR spectrophotometer. Food and Bioprocess Technology, 4(5), 809e813.
metric study with an example from near infrared spectroscopy. Journal of
Bao, Y., Liu, F., Kong, W., Sun, D., He, Y., & Qiu, Z. (2014). Measurement of soluble
Applied Spectroscopy, 54, 413e419.
solid contents and pH of white vinegars using VIS/NIR spectroscopy and least
Peng, B., Yue, T., & Yuan, Y. (2008). A fuzzy comprehensive evaluation for selecting
squares support vector machine. Food and Bioprocess Technology, 7(3), 54e61.
yeast for cider making. International Journal of Food Science Technology, 43(1),
Blanco, M., Peinado, A. C., & Mas, J. (2004). Analytical monitoring of alcoholic
140e144.
fermentation using NIR spectroscopy. Biotechnology and Bioengineering, 88,
Pouliot, M., Paquin, P., Martel, R., Gauthier, S. F., & Pouliot, Y. (1997). Whey changes
536e542.
during processing determined by near infrared spectroscopy. Journal of Food
Bock, J. E., & Connelly, R. K. (2008). Innovative uses of near-infrared spectroscopy in
Science, 62(3), 475e479.
food processing. Journal of Food Science, 73(7), 91e98.
Reeves, J. B. (2001). Near-infrared diffuse reectance spectroscopy for the analysis
Botelho, B. G., Mendes, B. A. P., & Sena, M. S. (2013). Development and analytical
of poultry manures. Journal of Agricultural and Food Chemistry, 49, 2193e2197.
validation of robust near-infrared multivariate calibration models for the
S.A.C.. (2008). GB/T 15038.Analytical methods of wine and fruit wine (pp. 20e41).
quality inspection control of mozzarella cheese. Food Analytical Methods, 6,
China Standards Press.
881e891.
Satora, P., Tarko, T., Sroka, P., & Blaszczyk, U. (2014). The inuence of Wick-
Buratti, S., Ballabio, D., Giovanelli, G., Dominguez, C. M. Z., Moles, A., Benedetti, S.,
erhamomyces anomalus killer yeast on the fermentation and chemical compo-
et al. (2011). Monitoring of alcoholic fermentation using near infrared and mid
sition of apple wines. FEMS Yeast Research, 14, 729e740.
infrared spectroscopies combined with electronic nose and electronic tongue.
Shenk, J. S., & Westerhaus, M. O. (1996). Calibration the ISI way. In A. M. C. Davies, &
Analytica Chimica Acta, 697, 67e74.
P. Williams (Eds.), Near infrared spectroscopy: The future waves (pp. 198e202).
Chen, Q., Ding, J., Cai, J., Sun, Z., & Zhao, J. (2012). Simultaneous measurement of
Chichester: NIR Pubilications.
total acid content and soluble saltfree solids content in Chinese vinegar using
Shen, F., Niu, X., Yang, D., Ying, L. B., Zhu, G., & Wu, J. (2010). Determination of amino
near-infrared spectroscopy. Journal of Food Science, 77(2), 222e227.
acids in Chinese rice wine by fourier transform near-infrared spectroscopy.
Chen, Q., Zhao, J., Fang, C., & Wang, D. (2007). Feasibility study on identication of
Journal of Agricultural and Food Chemistry, 58, 9809e9816.
green, black and oolong teas using near-infrared reectance spectroscopy based
Valderrama, P., Braga, J. W. B., & Poppi, R. J. (2007). Variable selection, outlier
on support vector machine (SVM). Spectrochimica Acta A, 66, 568e574.
detection, and gures of merit estimation in a partial least-squares regression
Cozzolino, D., Parker, M., Dambergs, R. G., Herderich, M., & Gishen, M. (2006).
multivariate calibration model. A case study for the determination of quality
Chemometrics and visible-near infrared spectroscopic monitoring of red wine
parameters in the alcohol industry by near infrared spectroscopy. Journal of
fermentation in a pilot scale. Biotechnology and Bioengineering, 95, 1101e1107.
Agricultural and Food Chemistry, 55, 8331e8338.
Di Egidio, V., Olivieri, P., Woodcock, T., & Downey, G. (2011). Conrmation of brand
Williams, P. C. (2001). Implementation of near-infrared technology. In
identity in foods by near infrared transectance spectroscopy using classica-
P. C. Williams, & K. H. Norris (Eds.), Near infrared technology: In theagricultural
tion and class-modelling chemometric techniques-the example of a Belgian
and food industries (pp. 145e171). St. Paul: American Association of Cereal
beer. Food Research International, 44, 544e549.
Chemists, Inc.
Di Egidio, V., Sinelli, N., Giovanelli, G., Moles, A., & Casiraghi, E. (2010). NIR and MIR
Wilson, S. M., Maguer, M. L., Duitschaever, C., Buteau, C., & Allen, O. B. (2003). Effect
spectroscopy as rapid methods to monitor red wine fermentation. European
of processing treatments on the characteristics of juices and still ciders from
Food Research and Technology, 230, 947e955.
Ontario-grown apples. Journal of the Science of Food and Agriculture, 54,
Giovenzana, V., Beghi, R., & Guidetti, R. (2014). Rapid evaluation of craft beer quality
215e224.
during fermentation process by vis/NIR spectroscopy. Journal of Food Engi-
Xie, L., Ye, X., Liu, D., & Ying, Y. (2011). Prediction of titratable acidity, malic acid, and
neering, 142, 80e86.
citric acid in bayberry fruit by near-infrared spectroscopy. Food Research Inter-
Grassi, S., Amigo, J. M., Lyndgaard, C. B. G., Foschino, R., & Casiraghi, E. (2014). Beer
national, 44(7), 2198e2204.
fermentation: monitoring of process parameters by FTNIR and multivariate
Xu, L., Zhou, Y., Tang, L., Wu, H., Jiang, J., Shen, G., et al. (2008). Ensemble pre-
data analysis. Food Chemistry, 155, 279e286.
processing of near-infrared (NIR) spectra for multivariate calibration. Analytica
Huang, H., Yu, H., Xu, H., & Ying, Y. (2008). Near infrared spectroscopy for on/in-line
Chimica Acta, 616(2), 138e143.
monitoring of quality in foods and beverages: a review. Journal of Food Engi-
Ye, M., Yue, T., Yuan, Y., & Li, Z. (2014). Application of FT-NIR spectroscopy to apple
neering, 87(3), 303e313.
wine for rapid simultaneous determination of soluble solids content, pH, total
Inon, F. A., Llario, R., Garrigues, S., & de la Guardia, M. (2005). Development of a PLS
acidity, and total ester content. Food and Bioprocess Technology, 7, 3055e3062.
based method for determination of the quality of beers by use of NIR spectral
Zhang, A., Fang, Y., & Li, X. (2011). Occurrence and estimation of trans-resveratrol in
ranges and sample-introduction considerations. Analytical and Bioanalytical
one-year-old canes from seven major Chinese grape producing regions. Mole-
Chemistry, 382, 1549e1561.
cules, 6, 2846e2861.
Kennard, R. W., & Stone, L. A. (1969). Computer aided design experiments.