Journal of Food Measurement and Characterization

https://doi.org/10.1007/s11694-021-01023-4

ORIGINAL PAPER

Detection of adulteration of kudzu powder by terahertz time‑domain

spectroscopy
Li Bin1 · Luo Bing2 · Liu Yan‑de1 · Wu Jian1

Received: 19 February 2021 / Accepted: 6 May 2021

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2021

Abstract
In this paper, the terahertz time-domain spectroscopy (THz-TDS) combined with chemometric methods, is used to explore
the rapid and undamaged detection of kudzu powder’s adulteration. The spectral data in the range of 0.1–7 THz is collected.
The absorption coefficient of the 0.5–3.0 THz band is selected for analysis. The chemometric methods are adopted to build
the analytical model of the spectrum’s absorption coefficient. In qualitative modelling, the partial least squares decision
analysis (PLS-DA) analyses the absorption coefficient data. The result shows that the misjudgment rate is 57.6%. The least
squares support vector machines decision analysis (LS-SVM-DA) basing on Linear (LIN), and radial basis function (RBF)
as the kernel function is used to analyze the data, the results show that the misjudgment rate is all 13.3%. Finally, the suc-
cessive projections algorithm (SPA) and the elimination of uninformative variables (UVE) are used to select the spectrum’s
absorption coefficient to optimize the model, the misjudgment rates of PLS-DA model basing on SPA and UVE are 33.3%
and 23.3%, respectively. The LIN of LS-SVM-DA and the RBF of LS-SVM-DA basing on SPA are all 43.3%, the misjudg-
ment rates of the LIN of LS-SVM-DA and the RBF of LS-SVM-DA basing on UVE are13.3% and 6.7%, respectively. The
comparison found that the RBF of LS-SVM-DA basing on the UVE prediction model for the adulteration of kudzu powder
is more accurate than the others. It can come true the judgment of adulteration of kudzu powder well. Studies have shown
that THz-TDS combined with chemometric methods can provide a fast and accurate analytical way for qualitative analysis
of kudzu powder’s adulteration.

Keywords Terahertz time-domain spectroscopy · Arrowroot · Partial least squares · Least squares support vector machines

Introduction sore throat, cold and fever, and skin inflammation. It can also
prevent some cardiovascular diseases such as hypertension,
Kudzu powder, which can be brewed with water, is natural hyperlipidemia, coronary heart disease, etc. Its medicinal
nutrition extracting from a vine named Pueraria lobata. Its and edible value is very high, known as ‘Asian Ginseng’
appearance is a generally dark white powder. Kudzu pow- [1–3]. However, due to the high price of pure kudzu pow-
der has the effects of clearing away heat and detoxification, der, there is a phenomenon of adulteration in the market. To
nourishing nutrition, preventing heatstroke, nourishing skin reduce the production cost, many illegal businessmen add
and skincare, and removing body garbage. It also has appar- other kinds of low-cost starch (such as lotus root powder,
ent curative effects on wind fire toothache, light mouth ulcer, potato powder, rice powder, etc.) into the pure kudzu pow-
der, to replace the inferior quality with the false, and sell
these adulterated kudzu powder to consumers at the price
* Li Bin of ordinary pure kudzu powder to earn the profit of cost
[email protected] difference [4]. This kind of behavior deceives consumers’
1
National and Local Joint Engineering Research Center money and loses kudzu powder’s significance as a health
of Fruit Intelligent Photoelectric Detection Technology care product.
and Equipment, Institute of Optical‑Electro‑Mechatronics At present, the identification techniques for adulterated
Technology and Application, East China Jiao Tong kudzu powder on the market include character comparison,
University, Nanchang 330013, China
microscope comparison, sensory identification, chromatog-
2
AVIC Hongdu Aviation Industry Group, Nanchang 330024, raphy liquid chromatography, near-infrared spectroscopy,
Jiangxi, China

13
Vol.:(0123456789)
 L. Bin et al.

Fourier transform infrared spectroscopy [5–7]. However, The surface of the sample section should be smooth without
these methods have their defects, such as character com- cracks.
parison method and microscope method have a cumbersome
operation, and low detection accuracy, The detection of sen- Experimental apparatus
sory identification is not accurate, chromatographic liquid
chromatography needs high cost and cumbersome process, The samples’ spectral data were collected by the terahertz
near-infrared spectroscopy and Fourier transform infrared time-domain spectrometer of TAS7400TS which are devel-
spectroscopy have low detection accuracy. Therefore, THz- oped by Edwin Company of Japan. The spectral resolution
TDS is proposed to determine whether kudzu powder is of the spectrometer is 7.6 GHz, the measurement spectrum
adulterated. Compared with other detection technologies, it range is 0.1–7 THz, the sampling scanning speed is 8 ms, the
is widely used in food safety detection due to its advantages scanning times are set to 1024 times, and the measurement is
such as short detection period, easy operation, high preci- conducted in transmission mode. Terahertz time-domain spec-
sion, and no damage to the tested samples [8–13]. Xiu-yang trometer has strict requirements on the using environment. The
et al. used THz-TDS technology to qualitatively and quanti- laboratory is located in a constant temperature and humid-
tatively detect the adulterated potato powder in kudzu Pow- ity, with the ambient temperature of 23 degrees centigrade
der [14, 15]. However, Du Xiuyang et al. only carried out a and air humidity of 25%. When the instrument is started, the
rapid detection of kudzu powder laced with potato powder. compressor needs to be opened to fill the sample detection
The adulteration of kudzu powder often mixed with various space of terahertz time-domain spectrometer with dry air.
kinds of starch (such as lotus root powder, potato powder, This is because the moisture in the air will affect the detection.
rice powder, etc.). Each time the machine is turned on, it needs to be filled about
There is no rapid and accurate detection technology given 30 min of dry air so that the humidity in the sample detection
the adulterated problem of kudzu powder with various kinds space is lower than 5%, and then the instrument can start to
of starch in the market. THz-TDS is proposed to detect the collect the spectral.
adulteration of kudzu powder with different types and con-
centrations of impurities. Combining with the chemomet- Optical parameters
rics method, the optimal qualitative model of adulteration of
kudzu powder is established, which provided the theoretical The terahertz time-domain spectrum of the sample is collected
basis and experimental reference for the market to detect by THz-TDS device, and then the time-domain information is
kudzu powder’s adulteration THz-TDS. converted into the frequency-domain information by Fourier
transform. The data processing is carried out using the method
proposed by Dorney et al. to obtain the absorption coefficient
Experimental part and refractive index of the sample [15].
The formula is as follows:
Sample preparation
𝜃c
n= +1 (1)
fD
The kudzu powder, lotus root powder, potato powder and rice
powder used in the experiment are all obtained by purchasing { }
raw materials and grinding them. Now kudzu powder, lotus 2 4n
𝛼 = ln (2)
root powder, potato powder, rice powder are put into the dry- D A[n + 1]2
ing oven, the temperature is adjusted to 60 degrees centigrade
of drying oven, and they are dried for 24 h. According to the where, n is the refractive index of the sample, 𝜃 is the spec-
experimental design requirements which the adulterating per- tral phase, c is the propagation velocity of light, D is the
centages of potato powder, lotus root powder, rice powder and thickness of the sample, f is the spectral frequency, 𝛼 is the
various powders are 30%, 30%, 30%, 60%, respectively, kudzu absorption coefficient of the sample, and A is the spectral
powder, lotus root powder, potato powder and rice powder are amplitude.
mixed and are placed in the centrifugal tube. A vortex stirrer
is used to oscillate the mixed powders for 4 min to ensure uni- Modeling
form distribution. The sample preparation adopts the pressing
method, that is, an appropriate amount of sample powders are Modeling method
taken from the sample tube after mixing evenly, the critical die
presses the powders for 1 min under the pressure of 10 MPa, Partial least squares (PLS) is one of the most widely used
and then the sample powders are packed into a sample piece multivariate regression methods in spectral analysis. The
with a diameter of about 1 cm and a thickness of about 1 mm. model formula is as follows:

13
Detection of adulteration of kudzu powder by terahertz time‑domain spectroscopy

and prediction set is used to assess the modelling quality.

X = TE + ax (3)
The values of the standard deviation of the calibration set
and prediction set are closer to 0, the modelling is better, and
Y = UF + ay (4) the prediction ability is more vital. If the correlation coef-
where, the matrix T and U are the score matrix of X and ficient values between the correction set and prediction set
Y, respectively; the matrix E and F are the loading matrix are more relative to 1, the correlation between the correction
of X and Y, respectively; ax and ay are the residual matrix set and prediction set is higher.
introducing by partial least squares model method of X and
Y, respectively.
The linear regression are done between T and U, that is, the Qualitative analysis of THz‑TDS combined
score matrix of X and the score matrix of Y are exchanged, so with chemometrics
that the components of X are directly related to Y each other.
LS-SVM is an improvement of support vector machine Qualitative spectral analysis
(SVM). Compared with SVM, LS-SVM has short training
time, low computational complexity and strong generaliza- Terahertz spectrometer is used to detect kudzu powder’s
tion ability. LS-SVM uses the least square linear system as adulterated samples and pure kudzu powder in reasonable
loss function, and it solves a set of linear equations instead of temperature and humidity. The detection frequency range of
complex quadratic programming in SVM. In this paper, the terahertz spectrometer is 0.1 ~ 7 THz. Because the signal-
LIN kernel function and RBF kernel function are selected, and to-noise ratio of 0.1 ~ 0.5 THz and 3.0 ~ 7 THz frequency
their formula is as follows: spectrum is too low, the 0.5 ~ 3.0 THz internal spectrum is
selected as the research band spectrum. In this experiment,
K(Xi , Xk ) = Xi × Xk (5) we need to prepare pure kudzu powder and adulterated
kudzu powder, respectively, mixed with potato powder, rice
‖X − X ‖2 powder and lotus root powder. The mass fraction ratio and
K(Xi , Xk ) = exp(− ‖
i k‖
) (6) the number of samples are shown in Table 1. Each model
2
(2𝜎 ) prepared 24 samples, a total of 120 samples. Four spectra
where, Xi is the sample point, Xk is the kernel function center are collected from 24 pieces of each mass fraction, and the
point, 𝜎 2 is the kernel parameter. average values are taken for subsequent processing.
Figure 1 shows the absorption spectra of pure kudzu pow-
Variable selection method der, kudzu powder mixing with potato powder, kudzu pow-
der mixing with lotus root powder, kudzu powder mixing
With the research and application of PLS and other methods, with rice flour and kudzu powder mixing with various kinds
the correction model with more vital prediction ability and of powders. Because kudzu powder, potato powder and rice
better robustness can be obtained by selecting specific wave- powder are all composed of starch, fat and carbohydrate,
length or wavelength region or eliminating nonlinear variables the shape of the five samples’ absorption spectra is simi-
by particular strategies. Therefore, in establishing a calibra- lar. There is no apparent characteristic peak. It can be seen
tion model, it is necessary to choose spectral variables in the from Fig. 1 that the absorption coefficients increase with the
samples to develop a robust model. SPA is a forward cycle increase of frequency in the range of 0.5 ~ 3.0 THz. Among
selection method. It starts from a wavelength, computes the them, the absorption coefficient of kudzu powder mixing
projection on the unselected wavelength, and introduces the with rice flour is the highest, and that of kudzu powder
wavelength with the largest projection vector into the wave- mixing with potato flour is the lowest. Therefore, according
length combination. Each newly selected wavelength has the
least linear relationship with the previous one. UVE is a wave- Table 1  Mass fraction and number of samples of 5 samples
length selection method basing on PLS regression coefficient,
Sample Mass fraction of alum Number of
it spectral variables are chosen by using regression coefficient number samples
as a measure of wavelength importance.
1 Pure kudzu powder 24
Evaluation of the model 2 70% kudzu powder + 30% potato powder 24
3 70% kudzu powder + 30% lotus root powder 24
For qualitative detection, the error rate is used to evaluate the 4 70% kudzu powder + 30% rice powder 24
quality of modelling. For quantitative detection, the standard 5 40% kudzu powder + 20% lotus root 24
powde + 20% rice powder + 20% potato
deviation and correlation coefficient of the calibration set powder

13
 L. Bin et al.

200
pure kudzu powder
and 13.3%, respectively, those show that the selection of
180
kudzu powder +30% potato powder kernel function has little effect on the modelling develop-
kudzu powder +30% lotus root powder
kudzu powder +30% rice powder
ment in the LS-SVM-DA modelling method. Using PLS-DA
160 kudzu powder with three kinds of powders and LS-SVM-DA for qualitative analysis of five samples,
Absorption Coefficients/cm-1

140 the results show that PLS-DA can’t achieve qualitative dis-
crimination of adulteration of kudzu powder; LS-SVM-DA
120
modelling method can achieve acceptable qualitative iden-
100 tification of adulteration of kudzu powder, but there is still
80
a high misjudgment rate of 13.3%.
Aiming at the problems that PLS-DA and LS-SVM-DA
60 have high misjudgment rate in the qualitative determina-
40 tion of adulteration of kudzu powder, this section attempts
to use SPA algorithm to screen the sample wavelengths.
20
It can be seen from Fig. 3a that the root means a square
0 error of the spectral data of the experimental samples using
0.5 1 1.5 2 2.5 3.1
Frequency/THz
SPA decreases with the increase of the number of spectral
data variables. When the number of variables is 30, the root
means square error reaches the minimum value. However, if
Fig. 1  Absorption coefficient diagram
the number of variables is too high or too low, the accuracy
of the spectral data model established by SPA will be prob-
to the level of the spectral absorption curve can’t identify lematic, if the number of variables is too low, some useful
whether there is the adulteration of kudzu powder accurately. spectra will be lost, and if the data information is too high,
So, the chemometrics method is needed to be combined to it will lead to overfitting. Therefore, the optimal number of
determine whether there is an impurity in kudzu powder. variables is determined to be 20 after careful consideration.
Figure 3b is the scatter diagram of the model obtained by
Qualitative analysis of modeling combined PLS-DA after the variables selected by SPA. The meanings
with chemometrics of S1, S2, S3 and S4 are the same as those in the previous
section. According to Fig. 3b, the scatter points of type 1
Firstly, PLS-DA and LS-SVM-DA are used to analyze the and 3 samples are relatively scattered, and there is a lot of
qualitative modelling of five samples, respectively. And the overlap with the representatives of type 2; they spread points
random partition method is used to split the calibration and of examples of category 2, 4 and 5 are relatively concen-
prediction, the calibration set and prediction set are dis- trated, but there is still some overlap between models. Sta-
tributed according to the ratio of 3:1. In LS-SVM-DA, two tistics show that ten samples out of 30 pieces are misjudged;
different kernel functions (LIN and RBF) are selected for that is, the misjudgment rate is 33.3%. Figure 3c and d are
modelling and classification. Figure 2a is the scatter diagram the scatter plots of the model obtained using LIN and RBF
of the model modelled by PLS-DA, where S1 with a value kernel function, respectively, and the variables are selected
of 1.5 is the boundary between the first type of sample and by SPA. From the scatter diagram of Fig. 3c and d, it can
the second type sample; S2 with a value of 2.5 is the bound- be seen that the misjudgment rate of samples obtained by
ary line between the second and third samples; S3 with a using LIN and RBF kernel functions are 43.3% and 43.3%,
value of 3.5 is the boundary between the third and fourth respectively, this shows that the selection of kernel func-
samples, and S4 with a value of 4.5 is the boundary between tion has little effect on the modelling development in the
the fourth and fifth samples. It can be seen from Fig. 2a that LS-SVM-DA modelling method. The results show that the
the scatter points of the first four samples are relatively scat- SPA algorithm can improve PLS-DA accuracy, but it can’t
tered. There are more overlaps among the various types of improve LS-SVM-DA accuracy. And the best misjudgment
samples, and the scatter points of the fifth type of models error rate is still 33.3%, this shows it can’t achieve the quali-
are relatively concentrated. Still, they overlap with the fourth tative discrimination of adulteration of kudzu powder.
type of pieces, according to statistics, 17 elements out of 30 Given the SPA’s high misjudgment rate- PLS-DA and
parts are misjudged; that is, the misjudgment rate is 56.7%. SPS-LS-SVM-DA in the qualitative determination of adul-
Figure 2b and c are the scatter plots of the model obtained teration of kudzu powder, this section attempts to use the
by using LIN and RBF kernel functions after LS-SVM-DA UVE algorithm to screen the sample wavelengths. Fig-
modelling, respectively. From b and c of Fig. 2, it can be ure 3a shows the change of regression coefficient with
seen that the misjudgment rate of samples obtained by LS- wavelength points selected by UVE, T1 and T2 are regres-
SVM-DA using LIN and RBF kernel functions are 13.3% sion coefficient thresholds. When the wavelength point’s

13
Detection of adulteration of kudzu powder by terahertz time‑domain spectroscopy

5.5
calibration (a)
5 prediction
4.5 S4
4

Prediction value
3.5 S3
3

2.5 S2
2

1.5 S1
1

0.5
1 2 3 4 5
Actual value

5 5
actual value actual value
prediction value (b) prediction value (c)

4 4
Classes

Classes

3 3

2 2

1 1
0 5 10 15 20 25 30 0 5 10 15 20
Number of samples Number of samples

Fig. 2  Scatter plots diagram of arrowroot adulteration. a Scatter plots diagram of PLS-DA, b scatter plots diagram of LS-SVM-DA based on
LIN, c scatter plots diagram of LS-SVM-DA based on RBF

regression coefficient is between T1 and T2, the wave- that is, the misjudgment rate is 23.3%. Figure 3c and d
length point is eliminated. Anyway, the wavelength point are the scatter plots of the models obtained using LIN
is retained, T3 is the cut off line of wavelength variables, and RBF kernel functions, respectively. From the scatter
and that is, the total number of wavelength points is 341. plots of c and d in Fig. 3, it can be seen that the mis-
It can be seen from Fig. 3a that 131 wavelength variable judgment rate of samples obtained by using LIN and RBF
points are retained. Figure 3b is the scatter diagram of kernel functions are 13.3% and 6.7%, respectively, which
the model obtained by PLS-DA modelling after the vari- means that the model basing on RBF function has better
ables selected by UVE. The meanings of S1, S2, S3 and effect in the LS-SVM-DA modelling method. Through the
S4 are the same as those in the previous section. From qualitative modelling and analysis of five samples by PLS-
Fig. 3b, the scatter points of the first and fifth samples DA and LS-SVM-DA after selecting UVE variables, the
are relatively scattered, and there are more overlaps with results show that UVE algorithm improves the accuracy
another example; the scatter points of the second, third and of PLS-DA significantly and improves the accuracy of LS-
fourth samples are relatively concentrated, but there is still SVM-DA. The best misjudgment rate is only 6.7%; this
a small amount of overlap between the samples. According indicates that it can come true the qualitative identification
to statistics, seven models out of 30 pieces are misjudged; of adulteration of kudzu powder well (see Fig. 4).

13
 L. Bin et al.

selected variables: 20 (RMSE = 0.51325)

1.8 5.5
calibration (b)
(a) 5 prediction
1.6
4.5 S4
1.4 4

Prediction value
1.2
3.5 S3
RMSE

3
1
2.5 S2

0.8 2
1.5 S1
0.6
1
0.4 0.5
0 10 20 30 40 50 1 2 3 4 5
Number of variables included in the model Actual value

5 5
prediction value (c) prediction value (d)
actual value actual value
4 4

3 3
Classes

Classes

2 2

1 1

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Number of samples Number of samples

Fig. 3  Scatter plots diagram of arrowroot adulteration. a The func- c scatter plots diagram of LS-SVM-DA based on LIN, d scatter plots
tion between residual errors of arrowroot adulteration spectrum and diagram of LS-SVM-DA based on RBF
screening variable by SPA, b scatter plots diagram of SPA-PLS-DA,

Comparative analysis of models rate and the less RMSEP than other modelling methods.
Therefore, the UVE algorithm is used to eliminate the non-
In this study, PLS and LS-SVM are used to establish the information variables, and LS-SVM-RBF-DA establishes
qualitative model of kudzu powder adulteration. In LS- the qualitative model.
SVM, two kinds of kernel functions are used to establish
the qualitative model of kudzu powder adulteration. SPA
and UVE algorithms are used to select spectral variables, the Conclusion
Root Mean Square Error of Prediction (RMSEP) is used to
assess the robustness of the model, the RMSEP is less, and In this paper, the qualitative analysis of adulteration of
the robustness of the model is well. The results are shown kudzu powder is carried out basing on THz-TDS combined
in Table 2. It can be seen from the table that: the use of with chemometrics method, that is, pure kudzu powder,
UVE-LS-SVM-RBF-DA algorithm has better misjudgment kudzu powder mixed with potato powder, kudzu powder

13
Detection of adulteration of kudzu powder by terahertz time‑domain spectroscopy

UVE plot when 3-components are considered in the PLS model

5.5
60 calibration
(a) 5 prediction (b)
40 4.5 S4
T3
4

Prediction value
20 T2 3.5 S3
tv alue

3
0
2.5 S2
-20 T1 2
1.5 S1
-40
1
0.5
0 100 200 300 400 500 1 2 3 4 5
real variables - index - random variables Actual value
5 5
actual value actual value
prediction value (c) prediction value (d)

4 4
Classes

Classes

3 3

2 2

1 1
0 6 12 18 24 30 0 6 12 18 24 30
Number of samples Number of samples

Fig. 4  Scatter plots diagram of arrowroot adulteration. a The function c scatter plots diagram of UVE-LS-SVM-DA based on LIN, d scatter
between regression coefficient of arrowroot adulteration spectrum and plots diagram of UVE-LS-SVM-DA based on RBF
screening variable by UVE, b scatter plots diagram of UVE-PLS-DA,

Table 2  The result contrast of Method Misjudg- RMSEP Method Misjudg- RMSEP
variety models ment rate ment rate
(%) (%)

PLS-DA 57.6 0.619 SPA-LS-SVM-RBF-DA 43.3 0.450

LS-SVM-LIN-DA 13.3 0.344 UVE-PLS-DA 23.3 0.416
LS-SVM-RBF-DA 13.3 0.323 UVE-LS-SVM-LIN-DA 13.3 0.302
SPA-PLS-DA 33.3 0.503 UVE-LS-SVM-RBF-DA 6.7 0.217
SPA-LS-SVM-LIN-DA 43.3 0.481

with lotus root powder, kudzu powder with rice flour and in the range of 0.5 ~ 3.0 THz and the absorption curves are
kudzu powder with various kinds of powders are qualita- similar, the qualitative identification can’t be made by height
tively analyzed. Because there are no prominent character- characteristics of the absorption coefficient. In qualitative
istic peaks in the absorption spectra of five types of samples modeling, PLS-DA is used for modelling and analysis. The

13
 L. Bin et al.

result shows that the misjudgment rate of PLS-DA model is 6. C. Jia, L. Jia, Ma. Ya-qing et al., Rapid detection of kudzu starch
57.6%, which can’t realize the qualitative discrimination of adulteration by near-infrared diffuse reflectance spectroscopy.
Food Sci. 35(8), 133–136 (2014)
adulteration of kudzu powder. Then LS-SVM-DA is used 7. L. Yan-de, C. Meng-jie, H. Yong, Application of spectral diagno-
to model and analyze. The result shows that the misjudg- ses technology in determination of agricultural products quality.
ment rate of model prediction of LS-SVM-DA is 13.3%. J. East China Jiaotong Univ. 35(04), 1–7 (2018)
The model result is good, which can realize the qualitative 8. Seung Hyun Baek, Heung Bin Lim, Hyang Sook Chun, Detection
of melamine in foods using terahertz time-domain spectroscopy.
identification of adulteration of kudzu powder. Finally, SPA J. Agric. Food Chem. 62(24), 5403–5407 (2014)
and UVE algorithm are used to screen the spectral variables 9. Ge. Hong-yi, Jiang Yu-ying, Xu. Hao-hui et al., Identification of
to optimize the model. The results show that the LS-SVM- wheat quality using THz spectrum. Opt. Express 22(10), 12533–
DA modelling method combined with the UVE algorithm 12544 (2014)
10. I. Amenabar, F. Lopez, A. Mendikute, In introductory review to
is the best, and its misjudgment rate was 6.7%. The results THz non-destructive testing of composite mater. J. Infrared Mil-
show that THz-TDS can be used for qualitative analysis of lim. Terahertz Waves 34(2), 152–169 (2013)
adulteration of kudzu powder. This study provides a method 11. C. Ze-wei, Z. Zhuo-yong, Z. Ruo-hua et al., Application of tera-
reference for the application of THz-TDS in the food field. hertz time-domain spectroscopy combined with chemometrics to
quantitative analysis of imidacloprid in rice samples. J. Quant.
Spectrosc. Radiat. Transfer 167, 1–9 (2015)
12. F. Crista, I. Radulov, L. Crista, A. Berbecea, A. Lato, Influence
Funding The funded was provided by Innovative Research Group Pro- of mineral fertilization on the amino acidcontent and raw protein
ject of the National Natural Science Foundation of China (Grant No. of wheat grain. Food Agric. Environ. 10, 47–50 (2012)
31760344). 13. Wu. Jian, L.I. Yang-de, Li. Bin et al., Qualitative and quantitative
detection of glucose and sucrose in milk powder based on tera-
hertz spectroscopy. Spectrosc. Spectr. Anal. 39(08), 2568–2573
(2019)
References 14. Li. Bin, Du. Xiu-yang, Liu Yan-de et al., Detection research of
seed potato starch in kudzu using terahertz time-domain spectros-
1. Wei Shu-yong, Progress in Cerebral Protection of Pueraria lobata copy. Laser Optoelectron. Progress 56(20), 203001-1-203001–6
(Willd.) Ohwi (Fabaceae) and Puerarin. Food Sci. 36(7), 259–263 (2019)
(2015) 15. Liu Yan-de, Du. Xiu-yang, Li. Bin et al., Detection of benzoic
2. A.K. Maji, S. Pandit, P. Banerji, Pueraria tuberosa: a review on acid in Arrowroot using terahertz technology. Laser Optoelectron.
its phytochemical and therapeutic potential. Natural Prod. Res. Progress 56(4), 041101-1-041101–7 (2019)
28(23), 2111–2127 (2014)
3. Wei Shu-yong, Chen Yi, Xu. Xiao-yu, Progress on the pharmaco- Publisher’s Note Springer Nature remains neutral with regard to
logical research of puerarin: a review. Chin. J. Natural Med. 12(6), jurisdictional claims in published maps and institutional affiliations.
407–414 (2014)
4. C. Jia, Li. Jian-chao, C. Jia et al., Discrimination of kudzu starch
adulteration by fourier transform infrared spectroscopy. Food Sci.
32(8), 226–230 (2011)
5. Du. Shun Liang, Z.M. Wei et al., Research on identification meth-
ods of kudzu vine root starch. Chin. J. Health Lab. Technol. 22(5),
1151–1152 (2012)

13