Accepted Manuscript

Recent developments of green analytical techniques in analysis of tea’s quality and

nutrition

Quansheng Chen, Dongliang Zhang, Wenxiu Pan, Qin Ouyang, Huanhuan Li, Khulal
Urmila, Jiewen Zhao

PII: S0924-2244(15)00029-1
DOI: 10.1016/j.tifs.2015.01.009
Reference: TIFS 1622

To appear in: Trends in Food Science & Technology

Received Date: 26 July 2014

Revised Date: 22 January 2015
Accepted Date: 31 January 2015

Please cite this article as: Chen, Q., Zhang, D., Pan, W., Ouyang, Q., Li, H., Urmila, K., Zhao, J., Recent
developments of green analytical techniques in analysis of tea’s quality and nutrition, Trends in Food
Science & Technology (2015), doi: 10.1016/j.tifs.2015.01.009.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

1 Recent developments of green analytical techniques

2 in analysis of tea’s quality and nutrition

3 Quansheng Chen ∗, Dongliang Zhang, Wenxiu Pan, Qin Ouyang,

Huanhuan Li, Khulal Urmila, Jiewen Zhao

PT
4

5 School of Food & Biological Engineering, Jiangsu University, Xuefu Road 301,

RI
6 Zhenjiang 212013, P.R. China

U SC
AN
M
D
TE
C EP
AC

∗ Corresponding author at: Tel.: +86-511-88790318; fax: +86-511-88780201.

E-mail address: qschen@ujs.edu.cn (QS Chen).
 ACCEPTED MANUSCRIPT

7 Abstract

8 Public attention in tea quality and nutrition has increased significantly in recent decades,

9 due in part to changes in consumer behavior and the gradually increasing consumption

10 of tea. Demand for high quality tea products obviously requires high standards of

PT
11 quality assurance and process control; satisfying this demand in turn requires

RI
12 appropriate analytical tools for analysis of tea quality and nutrition. Green analytical

SC
13 tool, as an alternative to conventional analysis methods, has the desirable features in

14 terms of operating speed, ease-of-use, minimal or no sample preparation, and

15
U
avoidance of sample destruction. This paper reviews recent developed technologies,
AN
16 such as near infrared (NIR) spectroscopy, electronic tongue (E-tongue), electronic nose
M

17 (E-nose), computer vision, integration of multiple sensors, and latest research efforts to

18 assess tea quality and nutrition. Particularly, we have reviewed some relevant data
D

19 processing algorithms involved in each green analytical tool. Finally, we provided the
TE

20 technical challenges and outlook for the application of these green analytical
EP

21 technologies in analysis of tea quality and nutrition.

22 Keywords: green analytical tool; near infrared spectroscopy; electronic tongue;

C

electronic nose; computer vision; integration of multiple sensors; data processing

AC

23

24 algorithms; tea; quality and nutrition.

25
 ACCEPTED MANUSCRIPT

26 Abbreviations: 2NM, 2-norm method; ANN, artificial neural network; APaRP,

27 augmented partial residual plots; AVE, absolute value expression; BiPLS/FiPLS,

28 backward/forward interval partial least squares; BPNN, back-propagation neural

29 networks; bPSO, binary Particle Swarm Optimization; BP-MLP, back

PT
30 propagation-multilayer perceptron; CA, cluster analysis; CCD, charged coupled device;

RI
31 CDA, canonical discriminant analysis; CP, conducting polymer; C-SVM, C-support

32 vector machine; DCT, discrete cosine transform; DOSC, direct orthogonal signal

SC
33 correction; DPLS, discriminant partial least squares; DWT, discrete wavelet transforms;

U
34 EC, epicatechin; ECG, epicatechin gallate; EGC, epigallocatechin;EGCG,
AN
35 epigallocatechin gallate; electronic nose, E-nose; ERM, empirical risk minimization;

36 electronic tongue, E-tongue; FCM, fuzzy C means; FDA, Fisher discriminant analysis;
M

37 FF-MLP, feed forward multilayer perceptron; GA, genetic algorithm; GAPLS, genetic
D

38 algorithm partial least squares; GC-MS, gas chromatography-mass spectrometry;

TE

39 GLCM, grey level co-occurrence matrix; GNN, genetic neural network; HPLC, High

40 performance liquid chromatography; iPLS, Interval partial least squares; ISE,

EP

41 Ion-Selective Electrode; KNN, K-nearest neighbors; LAPV, large amplitude pulse

C

42 voltammetry; LDA, linear discriminant analysis; LDL, low density lipoprotein; LR,
AC

43 linear regression; LS-SVM, least squares-support vector machine; LVQ, learning

44 vector quantization; MC, mean centering; MD, mahalanobis distance; MDM,

45 mahalanobis distance method; MLAPV, multifrequency large amplitude pulse

46 voltammetry; MLP, multi-layer perceptron; MOS, metal oxide semiconductor;

47 MOSFET, metal oxide silicon field effect transistor; MSAVI2, modified second
 ACCEPTED MANUSCRIPT

48 soil-adjusted vegetation index; MSC, multiplicative scatter correction; MSDA,

49 multiple stepwise discriminant analysis; MVT, multivariate trimming; NAS, net

50 analyte signal; NIR, near infrared; OSC, orthogonal signal correction; PCA, principal

51 components analysis; PCR, principal component regression; PCs, Principal

PT
52 components; PLS, partial least squares; PNN, probabilistic neural network; RBF, radial

RI
53 basis function; RHM, resampling by the half-means method; ROI, the region of interest;

54 TDNN, time-delay neural network; UVE, uninformative variable elimination; WMCS,

SC
55 weighted multiple scatter correction; WT, wavelet transforms; XRSD, X residual

U
56 standard deviation; SAPV, small amplitude pulse voltammetry; S-G smoothing,
AN
57 Savitzky–Golay smoothing; SIMCA, soft independent modeling of class analogy;

58 SiPLS, synergy internal partial least squares; SITO, Dynamic social impact theory
M

59 based optimizer; SL, statistical learning; SMLR, stepwise multiple linear regression;
D

60 SNV, standard normal variate transformation; SOM, self-organizing map; SPA,

TE

61 successive projections algorithm; SRM, structural risk minimization; SVD, singular

62 value decomposition; SVM, support vector machine; SVMR, support vector machine
EP

63 regression; SVR, support vector regression.

C

64
AC
 ACCEPTED MANUSCRIPT

65 Introduction

66 Tea is known to be one of the most popular beverages in the world with a long history,

67 which has formed its own culture in a specific area (Soylak, Tuzen, Souza, Korn, &

68 Ferreira, 2007). It is extremely season-specific and climate dependent with different

PT
69 categories all over the world, which is manufactured by a series of processes producing

RI
70 a range of variants from green, nonfermented tea to black fermented tea. Of the

71 approximately 2.5 million metric tons of dried tea manufactured, only 20% is green tea

SC
72 and less than 2% is oolong tea (Graham, 1992). Thus we can find various tea in the

U
73 market, almost all of which are from the top leaves of the plant Camellia sinensis (N.
AN
74 Bhattacharyya, Seth, Tudu, Tamuly, Jana, Ghosh, Bandyopadhyay, Bhuyan, et al.,

75 2007). Tea also provides a dietaty source of biologically active compounds that help
M

76 prevent a wide variety of diseases, including catechins, flavonols, anthocyanin,

D

77 depsides etc., which contributes to its beneficial medicinal properties. Wherein,

TE

78 catechins are believed as an anti-inflammatory and neuroprotective agent, help regulate

79 food intake and have an affinity for cannabinoid receptors, which may suppress pain
EP

80 and nausea and provide calming effects (Mandel, et al., 2005). Other active components
C

81 such as enzymes, carbohydrate, vitamins and minerals, also exert an important effect on
AC

82 improving health and even curing diseases. A growing body of evidence suggests that

83 moderate consumption of tea may protect against several forms of cancer,

84 cardiovascular diseases the formation of kidney stones, bacterial infections, and dental

85 cavities (Trevisanato & Kim, 2000), and substantial references worldwide have

86 interpreted the role of tea in nutrition and disease (Bhoo Pathy, et al., 2010; Deka & Vita,
 ACCEPTED MANUSCRIPT

87 2011; Dominguez-Perles, Moreno, Carvajal, & Garcia-Viguera, 2011; Hsu, 2005;

88 Mukhtar & Ahmad, 2000; Setiawan, et al., 2001; Weisburger & Chung, 2002; Y. Zhou,

89 et al., 2009). Therefore, it is of great significance to analyze tea quality and nutrition

90 during tea processing in order to keep high quality of the commercialized tea products.

PT
91 In addition, owing to the high value of tea products, the adulteration issue often

RI
92 happens, and as a result, abominable fake brand of tea products flood the market and

93 illegal merchants capitalize on the nondescript counterfeits for huge profits.

SC
94 Consequently, it is also important to distinguish certified products from counterfeited

U
95 ones.
AN
96 However, tea quality and nutrition analysis is a complex problem due to the presence

97 of innumerable compounds and their unpredictable reaction mechanism, and numerous

M

98 national and international authorities are setting criteria for quality attributes (Airy,
D

99 1999). Wet chemical analysis and human panel test are the most elementary and
TE

100 commonly used conventional methods for tea analysis. For example, colorimetric

101 measurements and titration using potassium permanganate were used for measurement
EP

102 of total polyphenol content (Wang, Helliwell, & You, 2000; Zuo, Chen, & Deng, 2002).
C

103 But both of them are time-consuming, laborious, and non-green methods. Human panel
AC

104 test for assessing tea quality is carried out by some skilled ‘tea tasters’, thus the

105 practical evaluation of tea quality is entirely subjective, leading to inaccurate results

106 owing to adaptation, fatigue and state of mind (Yang, et al., 2006). In the last few

107 decades, increasing interest on the use of accurate screening techniques or instrumental

108 methods to determine quality characteristics of foods has been of great interest to the
 ACCEPTED MANUSCRIPT

109 food industry. These techniques are subjective, highly repeatable, reproducible and

110 environmentally friendly. But some of them are not purely instrumental methods, for

111 example, gas chromatography-mass spectrometry (GC-MS) and high performance

112 liquid chromatography (HPLC) are applied to determine the internal components in tea

PT
113 (Del Rio, et al., 2004; Pongsuwan, Bamba, Yonetani, Kobayashi, & Fukusaki, 2008),

RI
114 which involves expensive instrumentation and time-consuming sample preparation

115 using solvents as well as analysis (from a few to several minutes). Therefore, they are

SC
116 not suitable to be used or adopted by the industry for rapid analysis or online

U
117 monitoring of quality.
AN
118 In order to enable food industries to rapidly respond to the changing demands of both

119 consumers and the market, there is the immediate need of appropriate analytical tools
M

120 for tea quality and nutrition analysis. Green analytical (or clean analytical) techniques
D

121 is the use of analytical chemistry techniques and methodologies that reduce or
TE

122 eliminate the use or generation of feedstocks, products, by-products solvents, reagents

123 etc., some of which are hazardous to human health or the environment (de la Guardia,
EP

124 2010). These techniques are easy-to-use, and can be used in-line or at-line to obtain
C

125 results quickly. With the technological improvement in green analytical tools, there are
AC

126 an increasing number of researches with respect to the utilization of some advanced

127 green analytical tools in tea quality analysis; however, there has been no systematic

128 article survey on green analytical tools in tea quality analysis. Therefore, we intended to

129 give a general view of green analytical techniques that have been recently employed for

130 tea quality analysis, including NIR spectroscopy, E-tongue, E-nose, computer vision,
 ACCEPTED MANUSCRIPT

131 and data fusion technology by integrating multiple sensors. Besides, we focused more

132 on exploiting the corresponding data preprocessing and modeling algorithms in their

133 studies, and at the end of this review, we provided the technical challenges and future

134 outlook for these green analytical tools, and also a summary was given.

PT
135 Specific Green Analytical Techniques

RI
136 Near Infrared Spectroscopy

137 Near infrared (NIR) spectroscopy is a spectroscopic method that uses the near infrared

SC
138 region of the electromagnetic spectrum ranging from 780 nm to 2500 nm. It is based on

U
139 the principle that different chemical bonds in organic matter absorb or emit light of
AN
140 different wavelengths when the sample is irradiated. Organic matter in samples has

141 distinct spectral fingerprints owing to the specific vibrational frequencies of chemical
M

142 bonds, which are determined by the mass of the constituent atoms, the shape of the
D

143 molecule, the stiffness of the bonds, and the periods of the associated vibrational
TE

144 coupling. The most prominent absorption bands are associated with molecular overtone

145 and combination vibrations of some hydrogen-based functional groups, such as O-H,
EP

146 C-H, C-O, N-H etc. (Weyer, 1985; Workman Jr, 1996). More complicated molecules
C

147 have many bonds, and vibrations can also be conjugated, resulting in two possible
AC

148 vibration patterns: stretch vibration and bent vibration. Therefore, specific absorption

149 of organic molecules in NIR region can express the chemical composition of the

150 material being analyzed (Alishahi, Farahmand, Prieto, & Cozzolino, 2010; Rohman &

151 Man, 2010). Moreover, different modes in NIR spectroscopy measurement are

152 employed for different samples, and the commonly used modes are transmittance and
 ACCEPTED MANUSCRIPT

153 diffuse reflectance (Huang, Yu, Xu, & Ying, 2008). Liquid samples, such as tea

154 infusion, often adopt transmittance mode in glass or quartz chamber with different sizes;

155 while solid samples, like tea leaves or powder, usually use diffuse reflectance with

156 carrier accessories (Cen & He, 2007).

PT
157 Generally, the whole evaluation procedure consists of the following four steps: (1)

RI
158 spectral data acquisition; (2) spectral data preprocessing; (3) calibration models

159 developing with a set of samples; and (4) models validating with a set of independent

SC
160 samples. Organic molecules in samples have their specific absorption patterns in NIR

U
161 region, which indicates chemical composition of the analyzed materials. However, the
AN
162 molecular overtone, and combination bands observed in NIR region are typically very

163 broad, leading to complex spectra. Thus, it can be difficult to assign specific features to
M

164 specific chemical components. Multivariate calibration techniques are often employed
D

165 to extract useful information and eliminate redundant information (Alishahi,

TE

166 Farahmand, Prieto, & Cozzolino, 2010). Spectral data preprocessing should be

167 implemented initially in order to eliminate noise and baseline shift from instrument and
EP

168 background, wherein, S-G smoothing (Savitzky & Golay, 1964), first and second
C

169 derivative, SNV (Barnes, Dhanoa, & Lister, 1989), MSC (Isaksson & Næs, 1988), WT
AC

170 (Tian, Yuan, Liu, & Lu, 2003), OSC (Wolda, Anttia, Lindgrenb, & Öhmanc, 1998),

171 DOSC (Westerhuis, de Jong, & Smilde, 2001), and NAS (Lorber, 1986), are the most

172 commonly used preprocessing methods. In addition, efficient variables selection is a

173 very important task for further spectral data modeling. Recently, considerable efforts

174 have been geared towards developing efficient variables selection methods which can
 ACCEPTED MANUSCRIPT

175 objectively identify variables to single out the ones that contribute useful information,

176 and/or eliminate the ones that contain serious noises (Zou, Zhao, Povey, Holmes, &

177 Mao, 2010). The commonly used variables selection methods include iPLS, SiPLS,

178 GAPLS, UVE and SPA.

PT
179 For model calibration, many pattern recognition methods have been employed for

RI
180 qualitative and quantitative analysis with NIR spectra. At first, PCA, as an

181 unsupervised pattern recognition method, has been extensively applied, not only for

SC
182 giving cluster trend but also for data reduction. With it, the original correlated data can

U
183 be transformed into a series of orthogonal principle components, which will be the
AN
184 input of calibration models. Apart from PCA, CA is also an unsupervised pattern

185 recognition method, which just gives cluster trend of samples. There are also numerous
M

186 supervised pattern recognition methods, such as LDA, KNN, DPLS, SIMCA, PLS,
D

187 ANN and SVM, which can be divided into linear (LDA, KNN, DPLS and SIMCA) and
TE

188 nonlinear (ANN and SVM) methods. Table 1 lists the multivariate calibration methods

189 for NIR spectra analysis used in recent reports.

EP

190 [Here for Table 1]

C

191 Tea is among one of the most popular beverages in the world because of its numerous
AC

192 benefits for health, many researchers have dedicated themselves into the study of tea

193 quality evaluation using NIR spectroscopy. Tea is abundant in hydrogen organic groups,

194 making it possible to utilize NIR spectroscopy for qualitative or quantitative analysis.

195 Since 1980s, NIR spectroscopy has been applied to determine the moisture content,

196 theaflavin content and other quality attributes in tea (Hall, Robertson, & Scotter, 1988).
 ACCEPTED MANUSCRIPT

197 In recent years, with the development of computer science and chemometrics, the

198 available applications of NIR spectroscopy technique have become more popular and

199 have attracted more attention from researchers for tea analysis owing to its high

200 consumption. For example, epigallocatechin gallate (EGCG), epigallocatechin (EGC),

PT
201 epicatechin gallate (ECG) and epicatechin (EC) are the typical catechins in tealeaves.

RI
202 In our previous study, NIR spectroscopy technique together with PLS algorithm was

203 used for simultaneous measurement of these four catechins (EGCG, EGC, ECG and EC)

SC
204 contents in green tea, and the results were achieved as shown in Fig.1. The results

U
205 indicated that FT-NIR spectroscopy prediction has high correlation with the reference
AN
206 measurement of catechins (i.e. HPLC). This study reveals that NIR spectroscopy in

207 combination with chemometrics has a high potential in simultaneous measurement of

M

208 main catechins contents in green tea, and also other micro-chemical components (Chen,
D

209 Zhao, Chaitep, & Guo, 2009).

TE

210 [Here for Fig.1]

211 According to the referenced works from Table 2, it is evident that NIR spectroscopy
EP

212 with the advantage of rapid, nondestructive and multi-component determination, has
C

213 gained wide acceptance in tea analysis. It records spectra for both qualitative
AC

214 (discrimination of tea categories, varieties and geographical origins) and quantitative

215 (determination of important chemical components in tea and optimization of steaming

216 process conditions) analysis of tea with mathematical tools.

217 For spectral preprocessing, SNV, MSC, WT, MIN/MAX, derivative and

218 Savitzky-Golay are often used comparatively to choose the optimal one, and our
 ACCEPTED MANUSCRIPT

219 previous studies proved that SNV preprocessing method achieved the best performance.

220 It can be demonstrated that SNV preprocessing method can remove slope variation and

221 correct light scatter in spectra collection because of different particle sizes of dry tea

222 leaves samples (Chen, Zhao, & Lin, 2009). Moreover, offset correction and detrend are

PT
223 also used for spectral pretreatment in some articles but occasionally. While modeling,

RI
224 PCA, LDA, KNN, ANN, SVM, and SIMCA are mainly qualitative algorithms, and

225 PLS, iPLS, SiPLS, SVMR, and BP-ANN are often used for quantitative analysis.

SC
226 For qualitative analysis, PCA is an unsupervised pattern recognition method, which

U
227 indicates data trends in a visualizing dimensional space, and compresses data to obtain
AN
228 the most effective variables (Chen, Zhao, & Lin, 2009). After PCA, supervised pattern

229 recognition methods (i.e. LDA, KNN, ANN, SVM, SIMCA, etc.) are used for further
M

230 research with the effective variables, which can be divided into linear algorithms (i.e.
D

231 LDA, KNN, SIMCA, etc.) and nonlinear (i.e. ANN, SVM, etc.) pattern recognition
TE

232 methods. Based on the overview of all the classification methods, nonlinear tools often

233 achieve better results in contrast to linear tools in most occasions. Because the linear
EP

234 tools alone may not provide a complete solution to the complicated data classification,
C

235 and nonlinear method is stronger than linear method in the level of self-learning and
AC

236 self-adjust. In addition, SVM is often superior to other traditional nonlinear tools (e.g.

237 ANN tool) in terms of tea classification. Because many of traditional nonlinear tools

238 based on the empirical risk minimization (ERM) principle suffer difficulties with

239 generalization, producing models that can over-fit the data. The ‘best’ mode by training

240 often results in worse predictive result, in other words, the generalization of the model
 ACCEPTED MANUSCRIPT

241 is worse. The foundation of SVM embodies the structural risk minimization (SRM)

242 principle, which has shown to be superior to the ERM principle. SRM minimizes an

243 upper bound on the expected risk, as opposed to ERM that minimizes the error on the

244 training data (Chen, Zhao, & Lin, 2009). What’s more, outlier study has been

PT
245 mentioned in some articles, and it is often constructed after PCA. It is equally important

RI
246 during and after modeling to detect possible outliers, and usually Sammon’s mapping,

247 MD, XRSD, RHM and MVT methods are often used comparatively (Zhang, Luypaert,

SC
248 Pierna, Xu, & Massart, 2004).

U
249 For quantification, there are also linear (PLS, PCR) and nonlinear (SVMR, BP-ANN)
AN
250 algorithms. They are usually performed comparatively whether it is linear relation or

251 not in the research work, in order to choose the best one. Such comparisons have been
M

252 carried out practically keeping on mind that the performance might have been better
D

253 due to over fitting. However, the facts support the otherwise. When the model
TE

254 developed with NIR spectra was used for measurement of complicated compositions in

255 tea, nonlinear algorithms (SVMR, BP-ANN) sometimes show great advantages over
EP

256 the linear algorithms (PLS, PCR) recalling the performances of these methods. First,
C

257 nonlinear tools for modeling have a strong capacity of self-learning, and can ignore
AC

258 minor PCs by self-adjusting weights. Besides, tea is a natural product with a large

259 number of complicated flavonoids, which contain a great quantity of hydrogenous

260 bonds (i.e. C-H, O-H, S-H, N-H, etc.), and NIR spectroscopy is based on overtones and

261 combinations of fundamental vibrations corresponding to hydrogenous bonds. These

262 overtone and combination bands in the NIR region are typically very broad, leading to
 ACCEPTED MANUSCRIPT

263 complex spectra, and it can be difficult to assign specific features to specific chemical

264 components. Thus, linear regression tools may not provide a complete solution to such

265 nonlinear cases. Of course, we cannot give the authoritative answer about which

266 algorithm to use in some given cases. But in most cases, the dilemma, i.e. whether if the

PT
267 nonlinear tool could be used for modeling or not, can be diagnosed by the approach of

RI
268 augmented partial residual plots (APaRP). A quantitative numerical tool (run test) was

269 employed to calculate the degree of nonlinearity based on APaRP approach, and if the

SC
270 value of APaRP exceeded the critical value, a nonlinearity conclusion was credible

U
271 (Chen, et al., 2012).
AN
272 In summary, it is apparent that NIR spectroscopy has gained extensive utilization in

273 tea analysis, both classification and quantification. Numerous algorithms have been
M

274 developed for spectral preprocessing and model calibration; most of them have
D

275 achieved satisfactory performances. All the applications show great potential for the
TE

276 instrument applied in the industry.

277 [Here for Table 2]

EP

278 Electronic Tongue

C

279 Electronic tongue (E-tongue) is also called as an artificial tongue, or taste sensor, and it
AC

280 can be considered as an analytical instrument that artificially reproduce the taste

281 sensation (Escuder-Gilabert & Peris, 2010). This analytical instrument is a typical array

282 of chemical sensors coupled with chemometrics for processing and characterizing

283 complex samples. Sensors can generate analytically useful electric signals when

284 interact with samples (Vlasov, Legin, Rudnitskaya, Di Natale, & D'Amico, 2005), and
 ACCEPTED MANUSCRIPT

285 various types of sensors have been applied in E-tongue systems, such as potentiometry,

286 voltammetry, and impedance technique.

287 In potentiometric E-tongue, a potential across a working electrode is measured when

288 an equilibrium state is reached, corresponding to the state where the net current is equal

PT
289 to zero. The E-tongue system composed of several kinds of lipid-polymer membranes

RI
290 (PVC membranes with lipid derivatives), was first presented by Toko and co-workers

291 (Iiyama, Ezaki, Toko, Matsuno, & Yamafuji, 1995). Potentiometric sensors, including

SC
292 ion-selective electrodes (ISEs) and lipid/polymer membranes electrodes, are often used

U
293 for multicomponent analysis, and they still remain the most widely used type in
AN
294 E-tongue systems, especially ISEs with the advantages of operation principle, low cost,

295 simple set-up and easy fabrication etc. Nevertheless, the major drawbacks of
M

296 potentiometric sensors are their temperature dependence and the adsorption of solution
D

297 components that affect the membrane potential. Many commercially available systems,
TE

298 i.e., the Taste Sensing System from Anritsu (Atsugi, Japan), Astree II E-tongue Sensor

299 from Alpha MOS (Toulouse, France), and the Multiarray Chemical Sensor from
EP

300 McScience (Suwon, Korea), are based on potentiometric measurements so far (Ciosek
C

301 & Wroblewski, 2007).

AC

302 The voltammetric sensors used in E-tongue systems were described in detail by

303 (Fredrik Winquist, Wide, & Lundström, 1997). The principle of voltammetric E-tongue

304 is that a potential is applied to a working electrode, resulting in current when

305 redox-active species are reduced or oxidized on the electrode surface, which is then

306 measured as per article (Winquist, 2008). The apparatus has been extensively used in
 ACCEPTED MANUSCRIPT

307 analytical chemistry due to its advantages of high sensitivity and selectivity, high

308 signal-to-noise ratio, low detection limits and various modes of measurement (i.e.,

309 cyclic, stripping and pulse voltammetry). What’s more, the surface of the electrodes can

310 be modified with various chemo sensitive materials obtaining sensors of various

PT
311 sensitivity and selectivity towards a variety of species. However, their application is

RI
312 limited to redox-active substances, and temperature dependence and large surface

313 alteration will cause drifts in sensor response. In order to solve this problem, a

SC
314 shelf-polishing equipment has been put forward recently (Olsson, Winquist, &

U
315 Lundström, 2006).
AN
316 An E-tongue based on impedance technique was first described by (Riul,

317 Malmegrim, Fonseca, & Mattoso, 2003). The impedance measurement is often
M

318 considered as a slow unnecessarily accurate method when used for repeated and rapid
D

319 electrical measurements, and is more often used for preliminary characterization of
TE

320 complex electronic circuits (Pioggia, Di Francesco, Marchetti, Ferro, & Ahluwalia,

321 2007). While measuring, there is no need of active species and standard reference
EP

322 electrode, but the device will be troublesome in many practical applications as a
C

323 reliable reference, which is a critical issue in miniaturized sensor arrays.

AC

324 Taste of tea infusions is the influential attribute of sensory information, and it is not

325 only an important factor in assessing tea quality, but also an important reference in tea

326 classification. Polyphenols, amino acids, caffeine, and catechin are main taste

327 compounds, which are measured by some physical and chemical experiments. The

328 results show good precision, accuracy and reliability, but they are time-consuming,
 ACCEPTED MANUSCRIPT

329 destructive and unsuitable for online monitoring. For example, in our previous study,

330 we used a commercial Astree II E-tongue Sensor together with chemometrics to

331 identify four different grades of tea (see Fig. 2). The results demonstrate that E-tongue

332 technology could be successfully used for classifying tea grades. E-tongue, as an

PT
333 emerging analytical tool, has a high potential in analysis of tea quality and nutrition

RI
334 (Chen, Zhao, & Vittayapadung, 2008).

335 [Here for Fig. 2]

SC
336 In order to observe the applications of this technique and the used algorithms,

U
337 substantial articles on tea analysis using E-tongue have been summarized as shown in
AN
338 Table 3, and most of which are about tea classification such as discrimination of tea

339 categories, varieties, geographical origins, grades, brands and fermentation degree. The
M

340 rest focuses on tea quantification, including the determination of caffeine, catechines,
D

341 sugar and amino acid L-arginine contents in tea.

TE

342 Before modeling, the data obtained from the E-tongue needs to be preprocessed.

343 There are several preprocessing techniques like auto scale, baseline subtraction,
EP

344 background subtraction, range scale1, range scale2, relative scale, SNV, WT, DWT etc.,
C

345 among which, auto scale and baseline subtraction preprocessing techniques are
AC

346 superior to the others (Bhattacharyya, et al., 2012; Palit, Tudu, Bhattacharyya, et al.,

347 2010). The experiments demonstrated that the combination of auto scale and baseline

348 subtraction showed excellent potential for preprocessing. In addition, DWT was used to

349 compress the exceptionally high number of response vectors, and WT was also

350 developed for the analysis of non-stationary signals as it can provide the time as well as
 ACCEPTED MANUSCRIPT

351 frequency information of a signal simultaneously (Palit, Tudu, Bhattacharyya, et al.,

352 2010).

353 When talking about modeling, PCA, LDA, KNN, SIMCA and ANN are the most

354 commonly used pattern recognition methods. E-tongue system is the application of the

PT
355 array of seven non-specific chemical sensors, which are cross sensitive towards

RI
356 different compositions in tea infusion, but they bring much overlapped information.

357 Thus, PCA, as an unsupervised pattern recognition method, is always used to extract

SC
358 maximum information from a very large amount of data and eliminate the overlapped

U
359 information (Chen, Zhao, & Vittayapadung, 2008). CDA, also stepwise LDA based on
AN
360 Fisher rule, is a useful component to PCA, and it was applied in some studies to assess

361 the adequacy of tea classification with satisfactory results (Xiao & Wang, 2009).
M

362 Besides, LDA, KNN, SIMCA, and ANN, as the supervised pattern recognition
D

363 methods, are often used comparatively and the results demonstrate that the nonlinear
TE

364 recognition methods are superior to linear recognition methods owing to its strong

365 capacity of self-learning and ignoring the minor PCs by self-adjusting weights.
EP

366 Moreover, bPSO, GA and PCR have also been used for classification under certain
C

367 circumstances, and a novel approach of comparing SITO with bPSO and GA has been
AC

368 attempted. Ultimately, SITO proved to be advantageous due to its spatially distributed

369 population and very few control parameters (Bhondekar, Kaur, Kumar, Vig, & Kapur,

370 2011). Another discrimination method by means of E-tongue is the application of

371 different waveforms like LAPV, SAPV and staircase voltammetry, and experiments

372 indicate that the best separation is obtained when staircase and LAPV are merged
 ACCEPTED MANUSCRIPT

373 (Ivarsson, Holmin, Hojer, Krantz-Rulcker, & Winquist, 2001). In addition, mechanized

374 flow technique-mainly flow injection strategy, is efficient to conquer the electrode

375 fouling problem, which will improve reproducibility for a long-term potential

376 measurement, and it has been applied for estimation of tea fermentation degree (Nieh,

PT
377 et al., 2009). With respect to tea quantification, PLS, PCR and PCA-ANN are used

RI
378 comparatively, and the results show that PCA-ANN is the optimal method due to its

379 stronger self-learning and self-adjusting capacities.

SC
380 Overall, E-tongue has achieved extensive applications in tea classification but a few

U
381 in tea quantification. Therefore, there is a need for further research in tea quantification
AN
382 and exploration of the appropriate algorithms for tea analysis. Thus, this technique

383 opens up new avenues in taste sensing and can be successfully implemented in the near
M

384 future for tea analysis.

D

385 [Here for Table 3]

TE

386 Electronic Nose

387 Developed in the early 1980s, the artificial olfaction (known as E-nose currently) has
EP

388 been widely used in diverse fields, such as agriculture, pharmaceutical, environmental
C

389 controls and clinical diagnostics (Persaud & Dodd, 1982). Generally, E-nose devices
AC

390 consisting of a sampling apparatus, a detector unit with a typical array of sensors and

391 data processing system for data recording and analysis, are designed to mimic the

392 mammalian sense of smell by producing a composite response unique to each odorant

393 (Yu, Wang, & Wang, 2009). The sensor array simulates the receptors in the olfactory

394 epithelium, and usually has a broad responsiveness to different odorants as the odor
 ACCEPTED MANUSCRIPT

395 receptor.

396 At present, several major categories of sensor arrays including piezoelectric sensors,

397 electrochemical sensors, optical and calorimetric or thermal sensors, have been

398 involved in the development of E-noses. Wherein, electrochemical sensors containing

PT
399 MOS, CP and MOSFET sensors, which are capable of converting chemical quantities

RI
400 into electrical signals that is related to the concentration of specific particles, have

401 received widespread utilization. The first developed sensor array was the array of MOS

SC
402 sensors, which can detect 20 odors (Casalinuovo, Pierro, Coletta, & Francesco, 2006).

U
403 MOS sensors are conductive in nature, and their resistance decreases when subjected to
AN
404 the odor vapor molecules. The ideal sensors would exhibit reliability, robustness,

405 sensitivity, selectivity and reversibility, thus it is of utmost importance to select

M

406 appropriate sensors.

D

407 In contrast to the conventional flavor analysis techniques such as human taste panel
TE

408 and GC-MS technique, E-nose devices show their great superiority, which can be

409 demonstrated as follows. Firstly, human taste panel composed of some trained tasters
EP

410 who have developed the language of their own to describe various tea aroma attributes
C

411 generally performs tea aroma evaluation. However, the language is difficult for
AC

412 consumers to comprehend, and the sensory test results would be affected by

413 physiological and environmental conditions. Secondly, tea aroma can also be

414 determined according to analysis of its volatile organic compounds by GC-MS

415 technique, whilst human smell sensation is a complex nonlinear system with a high

416 degree of sensitivity in tea’s aroma attributes identification, thus GC-MS technique
 ACCEPTED MANUSCRIPT

417 cannot be compared to the human panel test (Chen, Zhao, Chen, Lin, & Zhao, 2011).

418 Therefore, the appearance of E-nose technology makes tea aroma evaluation

419 easy-to-use and cost-efficient.

420 As an important quality factor of tea, aroma depends upon the amount of volatile

PT
421 organic compounds and their ratios. E-nose can be used for tea tasting and quality

RI
422 monitoring during production. Compared with the conventional methods, it is an

423 increasingly reliable, fast, and robust technology. Over the last decade, numerous

SC
424 applications of E-nose in tea quality detection have been reported and many studies

U
425 have been dedicated to improve the capability. For instance, in our previous study, we
AN
426 used a developed E-nose system together with chemometrics to identify tea grades, the

427 sketch of which is shown in Fig. 3. Herein, four grades of green tea, which were
M

428 classified by the human panel test, were taken into account. The E-nose system with
D

429 eight metal oxide semiconductors gas sensors array was developed for data acquisition,
TE

430 and chemometrics such as PCA, KNN, ANN and SVM, were used for data processing.

431 The final results of this research work demonstrate that E-nose technology with
EP

432 chemometrics can be successfully used for classification of tea grades (Chen, et al.,
C

433 2011). In conclusion, E-nose as an emerging analytical tool, has a high potential in
AC

434 analysis of tea aroma.

435 [Here for Fig.3]

436 E-nose system is the application of the sensor array consisting of several sensors

437 which are found to be adequately sensitive to the volatile. The sensor responses are

438 related to the amount of bio-chemical volatiles that comes in contact with the
 ACCEPTED MANUSCRIPT

439 corresponding sensor surface (Bhattacharyya, Seth, Tudu, Tamuly, Jana, Ghosh,

440 Bandyopadhyay, & Bhuyan, 2007). Based on the previous researches, results are in

441 accordance with the statistics in Table 4, which show that researchers have paid most of

442 their attention to tea qualitative analysis, including classification of tea categories,

PT
443 varieties, grades, types and geographical origins, identification of tea storage times and

RI
444 discrimination of tea fermentation process etc. Almost all the articles emphasize on the

445 discrimination of tea grades using MOS sensors due to its superior features, such as

SC
446 high sensitivity, high stability, reversibility and reliability over a long period. However,

U
447 MOS sensors also have some limitations. Higher sensitivity of this sensor to moisture
AN
448 and water vapor, which results in essential noise from strong sensor outputs produced

449 due to the presence of water vapor in the headspace of any sample. Such overriding
M

450 effect of noise brings catastrophic consequence in terms of efficient pattern recognition
D

451 by parametric and nonparametric methods. Therefore, a novel and elegant sampling
TE

452 system based on illumination heating and physical raking has been proposed to enhance

453 the sensitivity of MOS sensors (Bhattacharya, Tudu, Jana, Ghosh, Bandhopadhyaya, &
EP

454 Saha, 2008). In addition, a novel and portable odor imaging sensor named colorimetric
C

455 sensor which is insensitive to moisture and water vapor due to the hydrophobicity of the
AC

456 sensor materials and sensors plate, have been applied for classification of tea categories

457 in our recent research, the principle schematic diagram is shown in Fig.4 (Chen, Liu,

458 Zhao, & Ouyang, 2013). Other than MOS sensors, there are a few articles related to

459 MOSFET sensors which can work at much lower temperature than MOS sensors

460 (James, Scott, Ali, & O’hare, 2005), but odorant reaction product must penetrate the
 ACCEPTED MANUSCRIPT

461 catalytic metal layer, which needs to maintain a hermetic seal for the chip’s electrical

462 connections in harsh environments (Nagle, Gutierrez-Osuna, & Schiffman, 1998).

463 Recently, a new sensors array called chemiluminescence sensors array based on

464 catalytic nanomaterials was developed that has proved its potential for discrimination

PT
465 with simple sensing elements and instrumentation as well as reversible response and

RI
466 long-term stability (Niu, 2012).

467 [Here for Fig. 4]

SC
468 After data acquisition from E-nose system, normalization was usually applied to

U
469 preprocess the signals prior to model calibration while classifying. Then, PCA was
AN
470 applied to extract useful information and eliminate the redundant information because

471 the eight non-specific sensors have cross-sensitivity which contributes to the presence
M

472 of collinear variables among eight characteristic variables (Chen, et al., 2011).
D

473 However, this method tends to be unsatisfactory due to its indifference to the relation of
TE

474 data points with the specified classes, which implies that a supervised classification

475 method is needed to solve this problem (Dominguez-Perles, et al., 2011). Therefore, an
EP

476 innovative data clustering approach was investigated by means of the combination of
C

477 PCA, FCM and SOM. Here, PCA tries to estimate the most suitable positions of the
AC

478 probable clusters; FCM tries to locate the most probable cluster center within the data

479 points whose points could be representative of a cluster for a particular group or class,

480 and SOM nodes of SOM network try to take positions of the probable clusters’ center

481 which represent the positions of cluster centers (Dutta, Hines, Gardner, Kashwan, &

482 Bhuyan, 2003). Subsequently, LDA, KNN, ANN (MLP, LVQ, PNN, RBF, and BP) and
 ACCEPTED MANUSCRIPT

483 SVM are the most commonly used approaches for model calibration, and some of them

484 have been performed comparatively in some articles. The overall results reveal that

485 SVM embodies the excellent robustness both in experiments and in theory. Although,

486 ANN would easily suffer from over fitting of data in some cases, which will finally

PT
487 bring a worse prediction result. it has been explained that the nonlinear classification

RI
488 tools (ANN, SVM) involves parameters optimization and iteration, so are superior to

489 those of linear classification tools (LDA, KNN) (Chen, et al., 2011; Yu & Wang, 2007).

SC
490 While comparing the four approaches of MLP, LVQ, PNN and RBF networks, we can

U
491 observe that RBF network demonstrated satisfactory performance in the aspect of
AN
492 incremental learning-plasticity and stability (Dutta, et al., 2003; Palit, Tudu,

493 Bhattacharyya, et al., 2010; Tudu, Jana, et al., 2009). In the past few years, there have
M

494 been some methods introduced which are not commonly used. For instance, conformal
D

495 predictors based on SVM has been investigated for multi-class classification problems
TE

496 and excellent performance has been achieved (Ilia, Guang, & Alexander, 2010). An

497 approach named bionic olfactory neural network is accurate in simulating the olfactory
EP

498 system, which shows good features of fewer training times and less training sets in
C

499 contrast to the conventional ANN model (Yang, et al., 2006). Additionally, only a few
AC

500 applications are related to quantitative analysis, such as the prediction of odor mixture’s

501 responses on machine olfaction sensors based on DWT and SVR with GA to search the

502 optimal parameter.

503 In conclusion, excellent performances and advantages of E-nose in tea analysis

504 applications illustrate an attractive and effective means to replace the traditional tea
 ACCEPTED MANUSCRIPT

505 assessment methods. Nevertheless, further work is needed to assess the long-term

506 sensitivity, selectivity, and reliability of the E-nose system as for the quantitative

507 analysis of tea and the performance optimization of the devices.

508 [Here for Table 4]

PT
509 Computer Vision

RI
510 Computer/machine vision systems are gaining high popularity in the food industry with

511 its origin traced back to the 1960s (Patel, Kar, Jha, & Khan, 2011). Computer vision is

SC
512 an engineering technology that combines mechanics, optical instrumentation,

U
513 electromagnetic sensing, digital video and image processing technology together. A
AN
514 computer vision system generally includes: illumination, a camera, an image capture

515 board(frame grabber or digitizer), computer hardware and software (Brosnan & Sun,
M

516 2004). Typically, charged coupled device (CCD camera) is often used in modern vision
D

517 systems as the image sensors.

TE

518 After the image collection by computer vision, a series of image processing are used

519 to enhance and improve the acquired images for further analysis. Dilation: to increase
EP

520 brightness of each pixel surrounded by neighbors with a higher intensity. Threshold: to
C

521 convert the grey scale image into binary image. Spatial filters: to improve the image
AC

522 quality by removing noise and smoothing, sharpening, and transforming the image.

523 Grayscale morphology: to extract and alter the structure of particles in an image.

524 Frequency-domain processing: to remove unwanted frequency information (Patel, et al.,

525 2011). Color and texture are important physical parameters to determine the quality of

526 tea but these parameters cannot be tackled using the traditional quality evaluation
 ACCEPTED MANUSCRIPT

527 processes, which are non-standardized and are based on subjective judgments made by

528 highly trained human sensory panel. For better results, image-processing techniques

529 have been applied preferentially to determine these parameters. Among them, various

530 techniques for color analysis include determination of color either colorimetrically or

PT
531 spectrophotometrically, and texture features including uniformity, density, coarseness,

RI
532 roughness, regularity, linearity, directionality, direction, frequency and phase are

533 usually extracted as variables. With these features, classification can be realized using

SC
534 statistical methods, geometrical, model-based and transform-based methods (Gill,

U
535 Kumar, & Agarwal, 2011). Further, if research efforts can be oriented towards the
AN
536 establishment of the correlation of physiochemical results with color and texture

537 profiles, image analysis could yield significant effects. Recently, novel imaging
M

538 technologies, such as hyperspectral imaging and multi-spectral imaging, have been
D

539 widely applied for tea analysis, and the systems can be described as follows.
TE

540 Hyperspectral imaging is an emerging technique that integrates conventional

541 imaging and spectroscopy to attain both spatial and spectral information from an object,
EP

542 and the device system typically contain the following components: objective lens,
C

543 spectrograph, camera, acquisition system, translation stage, illumination, and computer.
AC

544 Hyperspectral images, known as hypercubes (Lu & Chen, 1998), are three-dimensional

545 blocks of data, comprising of two spatial and one wavelength dimension (Gowen,

546 Odonnell, Cullen, Downey, & Frias, 2007). The hypercube, as shown in Fig. 5, allows

547 for the visualization of biochemical constituents of a sample since regions of a sample

548 with similar spectral properties have similar chemical composition, and obtaining
 ACCEPTED MANUSCRIPT

549 surface feature information according to the hyperspectral images. After hyperspectral

550 image acquisition, spectral preprocessing, variables selection, image extraction and

551 processing are carried out for analysis.

552 [Here for Fig.5]

PT
553 Multi-spectral imaging functions in different filters fitted to images, and images of

RI
554 specified wavelength can be collected. The multi-spectral imaging system applied to

555 assess tea quality typically consists of a three-channel common aperture camera, a

SC
556 frame grabber, several tungsten halogen light, and a computer. Multi-spectral imaging

U
557 is developed based on hyperspectral imaging using the selected wavelengths, thus it is
AN
558 advantageous than hyperspectral imaging due to the decrease of huge data (Li, He, Qiu,

559 & Wu, 2007).

M

560 In summary, computer vision provides an automated, non-destructive, and

D

561 cost-effective technique to accomplish inspection and evaluation in industry, which has
TE

562 been reported in recent decades. This approach based on image analysis and processing

563 has provided a great potential in tea quality assessment. Visual attributes are important
EP

564 quality parameters of tea, and major physical quality attributes are color, grain shape
C

565 and size, which are usually determined by organoleptic methods such as human sensory
AC

566 panel and analytical instruments. However, these methods are time-consuming,

567 laborious, expensive, inconsistent, and variable. Thus, in order to overcome these

568 shortcomings, the technology of computer vision aims to duplicate the human vision

569 via electronic perception. The understanding of an image provides the alternative for an

570 automated, non-destructive, and cost-effective technique to evaluate and discriminate

 ACCEPTED MANUSCRIPT

571 tea. Normal camera imaging, hyperspectral imaging, and multispectral imaging are the

572 commonly used computer vision tools.

573 In accordance to the reviewed articles listed in Table 5, we can observe that most of

574 the studies relate to the qualitative analysis (i.e. classification of tea), including

PT
575 discrimination of tea categories, varieties, grades, brands and colors and classification

RI
576 of tea grains under different illumination conditions. Only a few researches are about

577 the quantification of tea such as determination of chlorophyll content and distribution

SC
578 in tea plant’s leaves. Among them, discrimination of tea categories and grades seem to

U
579 be the most popular. Principal components as the useful information rather than
AN
580 redundant information are often extracted before modeling, and then as the input of the

581 classification model. By the same token, a novel method of wavelet texture analysis
M

582 assembled with GLCM was proposed for texture feature extraction (Li, He, & Qiu,
D

583 2008). LDA, SIMCA, ANN, SVM, and GNN are the most commonly used model
TE

584 calibration algorithms for tea classification, and some of them were used comparatively

585 in some papers. The overall results show that SVM is very specialized for classification
EP

586 due to its beneficial theory over others (Li, He, Qiu, & Bao, 2008). A new method
C

587 combining the advantages of DCT and LSSVM was put forward to discriminate tea
AC

588 categories, and the result demonstrated that the proposed method can be efficiently

589 utilized for texture recognition (Wu, Chen, & He, 2009). Furthermore, C-SVM, MSDA,

590 and STEPDISC were also proposed in some articles and finally they proved to be

591 considerably feasible. With regard to quantitative analysis of tea, determination of

592 chlorophyll content and distribution in tea plant’s leaves have been studied with the
 ACCEPTED MANUSCRIPT

593 MSAVI2 model using the hyperspectral imaging technique, and as a result have

594 achieved excellent performance (Zhao, Wang, Ouyang, & Chen, 2011).

595 These studies present the recent developments and applications of computer vision in

596 tea analysis, which indicates that computer vision has the potential to become a vital

PT
597 component of automated tea processing. This can be ascribed to that the increased

RI
598 computer capabilities and greater processing speed of algorithms are continually

599 developing to meet the necessary of online production. We anticipate that computer

SC
600 vision will achieve widespread utilization of the emerging platform technology in the

U
601 future.
AN
602 [Here for Table 5]

603 Integration of Multiple Sensors

M

604 The great advance of the so-called chemical sensors and their growing applications in
D

605 food analysis have attracted a considerable research interest in developing alternative
TE

606 techniques or new applications in the field of food sensory analysis. Accordingly, most

607 of green analytical tools as mentioned above, including NIR spectrograph, E-nose,
EP

608 E-tongue, and computer vision, have been adopted as alternative instruments in tea
C

609 analysis.
AC

610 In general, tea quality contains various aspects, such as taste, aroma, appearance and

611 flavor (Kovács, et al., 2010). Human panel test, as a commonly used methodology,

612 makes use of multiple mammal sensory organs (i.e. eye, nose, and tongue) to perceive

613 the various aspects of tea quality (i.e., color, aroma, and taste), and then mammal brain,

614 in accordance with the prior experience, can give a final decision to comprehensive
 ACCEPTED MANUSCRIPT

615 quality of tea (Nazifah, et al., 2011). Accordingly, the artificial senses of various

616 sensors including E-nose, E-tongue, computer vision (E-eye), are arranged to the

617 structures that are similar to the human senses. The analysis carried out by each green

618 analytical instrument provides data of tea samples, which can be related to the sensory

PT
619 attributes that allows characterization with suitable chemometrics. However, the

RI
620 complete characterization of tea samples requires simultaneous use of several

621 techniques that can describe the gustatory, olfactory and visual aspects. The

SC
622 comprehensive data coming from different sensors provides complementary

U
623 interpretations and facilitates a full description. The combination of information
AN
624 provided by several analytical instruments is called data fusion. It involves treating a

625 large number of multivariate signals from different sensors, and increases the use of
M

626 chemometrics.
D

627 There includes three levels at which the sensor data can be fused. The most basic, the
TE

628 so-called low-level fusion, consists of building a single data vector with the fused

629 signals from different sensors to form a sort of supra-sensor system. Next, the
EP

630 medium-level fusion consists of extracting features from the signals of each sensor and
C

631 then fusing them. Finally, in the high-level fusion, classification models are constructed
AC

632 separately for each sensor technique, and then, the individual classification results are

633 combined to produce the final classification (Chen, Zhao, Cai, & Saritporn, 2008). The

634 use of data fusion strategies is motivated by the theory that the data coming from

635 multiple green analytical techniques would improve the performance of the

636 methodology in contrast to the data obtained by a single technique.

 ACCEPTED MANUSCRIPT

637 In our previous research work, we attempted to discriminate tea grades by combining

638 two green analytical technologies of NIR spectroscopy and computer imaging, the

639 sketch of which is shown in Fig. 6. Herein, four grades of tea were taken, data fusion

640 was carried out at the medium-level, and BPNN was used for modeling. The final

PT
641 results show that the performance of the combined green analytical methodology is

RI
642 superior to the single technique (Chen, Zhao, Cai, et al., 2008). Zakaria Ammar et al.

643 presented a study for tea classification by combining E-nose and E-tongue together

SC
644 with PCA and LDA (Zakaria, et al., 2010). The results demonstrated that data fusion

U
645 based on E-nose and E-tongue could improve the discrimination rate. For superior
AN
646 combination of the responses from different types of sensor systems, Nazifah et al.

647 (2011) developed a human sensory mimicking system combining E-nose and E-tongue,
M

648 and the experiments in identifying four types of tea demonstrated that
D

649 multiple-modality systems performed better than that with single modality (Nazifah, et
TE

650 al., 2011). Banerjee et al. tested black tea quality via the responses combination of

651 E-nose and E-tongue, resulting in clearer clustering compared to the individual systems
EP

652 (Banerjee, et al., 2012). These studies mentioned above have sufficiently demonstrated
C

653 the potential of data fusion methodology, which represents a start point for future
AC

654 research that involves studies aimed at developing an intelligent evaluation system of

655 tea. However, to perform a complete study and validate the results, the sensory

656 interpretations is necessary, i.e., the information provided by a human panel test

657 consisting of skillful tea panelists or by consumers.

658 [Here for Fig.6]

 ACCEPTED MANUSCRIPT

659 Technical Challenges and Outlook

660 Based on the reviewed references and observed trends, we summarize the recent

661 developments and point towards some of the potential future directions in the field.

662 Most of the articles in this review are those published since 2005. As seen in Tables 2-5

PT
663 and the examples from the data fusion technology by integrating multiple sensors, we

RI
664 can see that green analytical technologies have become valuable tools for the inspection

665 of tea quality and nutrition. In particular, the development of sensors technique,

SC
666 computer hardware processing power and data processing algorithm, and the

U
667 application of specific research developments were described in this review article.
AN
668 Some advanced green analytical tools have enabled the rapid and objective inspection

669 of tea quality and nutrition. Furthermore, most of the green analytical technologies
M

670 have become valuable throughout the entire value chain of the tea industries. The
D

671 general trend seems to aim at the use of these green analytical technologies to optimize
TE

672 production cost and to guarantee tea product quality in many aspects. Based on the

673 referenced works and the observed trends from Tables 2-5, we try to provide the
EP

674 technical challenges and outlook for these green analytical techniques in analysis of tea
C

675 quality and nutrition.

AC

676 As mentioned above, tea quality covers various aspects, for example, external factors

677 (i.e., color, gloss, and consistency), and internal factors (i.e., aroma, taste, chemical

678 compositions, and microbial). The single green analytical technology, with

679 specific-application, will face great challenges in analysis of tea quality and nutrition.

680 Therefore, cost-effective solutions to challenging analytical problems in the future will
 ACCEPTED MANUSCRIPT

681 have to make use of integration of multiple green analytical technologies and even

682 non-green analytical sensors. Accordingly, the ever-increasing demand for the success

683 of this technology is to explore and develop more advanced data fusion methodologies

684 and architectures. Intelligent evaluation, using integrated multiple green analytical

PT
685 technologies of E-eye (i.e. computer vision), E-nose, and E-tongue, will provide a

RI
686 future developing trend in tea quality inspection. However, the multiple mammal

687 perception organs, in human sensory evaluation might have cross-response to tea

SC
688 quality attributes. Therefore, how to achieve the best data-fusion from multiple sensors

U
689 will be a great challenge to the success of intelligent evaluation of tea quality, like
AN
690 multiple mammal perception. Besides, the choice of suitable nonlinear pattern

691 recognition for modeling is also of great importance to achieve the intelligent
M

692 evaluation of tea quality.

D

693 Multivariate calibration models play a key role in these green analytical
TE

694 measurements. Especially for NIR technique in quantifying the composition content in

695 tea, we have to build a multivariate calibration model using a set of training samples.
EP

696 However, there are various kinds of tea in the market because of different geographical
C

697 origins and manufactured processes (Al-Othman et al. 2012; Soylak et al. 2007).
AC

698 Accordingly, the reliability of the calibration model depended on a specified kind of

699 samples will cause problem when measuring the different kind of sample. Therefore,

700 we have to build a calibration model for each kind of tea in practical usage; alternatively,

701 we can collect as many as possible various kinds of tea samples for modeling to

702 improve the reliability of the model.

 ACCEPTED MANUSCRIPT

703 The major barrier for application of these advanced green analytical tools in tea

704 industry is the budget constraints. Most of these green analytical tools are still

705 nonviable in many potential applications for its unacceptable cost. For example, current

706 E-nose technology is only as the lab setup, because most of them are expensive

PT
707 commercial equipments with time-consuming data analysis process. To develop a

RI
708 cost-effective and cheap specific E-nose system, is especially a great engineering

709 challenge to the analysis of tea quality. Given that the data obtained from most of the

SC
710 green analytical tools is complex and non-specific for inspection of tea properties, the

U
711 next challenge is data interpretation and accurate of the data. Data processing speed is
AN
712 still a bottleneck in heavy-duty and real-time applications, i.e. failing to handle the large

713 data streams. Therefore, choosing an advanced and practical machine-learning

M

714 algorithm is a precondition for a successful application of these advanced green

D

715 analytical tools in analysis of tea quality and nutrition. On one hand, developing
TE

716 adequately efficient and accurate data processing algorithms can accelerate processing

717 speed to meet modern manufacturing requirements. On the other hand, integrating data
EP

718 processing algorithms onto specialized hardware can significantly reduce time
C

719 consumption. With cheap and fast solutions in software and hardware, data processing
AC

720 techniques will play an increasingly important role in the rapid and intelligent

721 inspection of tea quality and nutrition.

722 Most of green analytical technologies have the capabilities for the soft-sensing

723 application in tea processing industry. Automation of tea production will continue to

724 increase in terms of that most applications are not automated currently. As the online
 ACCEPTED MANUSCRIPT

725 monitoring of tea quality and nutrition using the green analytical technologies, together

726 with flexible automation in tea processing equipments, becomes available, there will be

727 a need for monitoring tea production in order to guarantee tea quality homogeneity and

728 reduce energy consumption of the processing equipment. In particular, the technologies

PT
729 of NIR spectroscopy and computer vision have the potential for real-time perceiving

RI
730 the change in the internal/external attributes during tea processing. Thus, it enables the

731 online process understanding and optimization of process parameters in the automation

SC
732 of tea process.

U
733 Conclusions
AN
734 During the past two decades, considerable efforts have been made to explore the

735 possibilities offered by instrumental methods or sensors to monitor and predict changes
M

736 related with aroma, taste and other internal or external properties of tea. In this review,
D

737 we summarized recent developments of green analytical techniques as well as

TE

738 corresponding data preprocessing and modeling algorithms in tea quality and nutrition

739 analysis, and also provided technical challenges and future outlook.
EP

740 Acknowledgements
C

741 The National Natural Science Foundation of China (31471646) financially supported
AC

742 this work. We are also grateful to many of our colleagues for stimulating discussion in

743 this field.

 ACCEPTED MANUSCRIPT

744 References
745 Airy, A. (1999). Sub-committee on ISO tea to meet in Calcutta next month.

746 Alishahi, A., Farahmand, H., Prieto, N., & Cozzolino, D. (2010). Identification of transgenic foods

747 using NIR spectroscopy: a review. Spectrochimica Acta Part A: Molecular and Biomolecular

748 Spectroscopy, 75, 1-7.

PT
749 Alishahi, A., Farahmand, H., Prieto, N., & Cozzolino, D. (2010). Identification of transgenic foods

750 using NIR spectroscopy: A review. Spectrochimica Acta Part a-Molecular and Biomolecular

RI
751 Spectroscopy, 75, 1-7.

752 Al-Othman, Z.A., Yilmaz, E., Sumayli, H.M., Soylak, M., 2012. Evaluation of trace metals in tea

SC
753 samples from Jeddah and Jazan, Saudi Arabia by atomic absorption spectrometry. Bulletin of

754 environmental contamination and toxicology, 89(6), 1216-1219.

755

U
Banerjee, R., Tudu, B., Shaw, L., Jana, A., Bhattacharyya, N., & Bandyopadhyay, R. (2012).
AN
756 Instrumental testing of tea by combining the responses of electronic nose and tongue. Journal

757 of Food Engineering, 110, 356-363.

M

758 Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal variate transformation and

759 De-trending of near-infrared diffuse reflectance spectra. Applied Spectroscopy, 43, 772-777.
D

760 Bhattacharya, N., Tudu, B., Jana, A., Ghosh, D., Bandhopadhyaya, R., & Bhuyan, M. (2008).
TE

761 Preemptive identification of optimum fermentation time for black tea using electronic nose.

762 Sensors and Actuators B: Chemical, 131, 110-116.

EP

763 Bhattacharya, N., Tudu, B., Jana, A., Ghosh, D., Bandhopadhyaya, R., & Saha, A. B. (2008).

764 Illumination heating and physical raking for increasing sensitivity of electronic nose
C

765 measurements with black tea. Sensors and Actuators B: Chemical, 131, 37-42.
AC

766 Bhattacharyya, N., Bandyopadhyay, R., & Bhuyan, M. (2008). Electronic nose for black tea

767 classification and correlation of measurements with Tea Taster marks. IEEE TRANSACTIONS

768 ON INSTRUMENTATION AND MEASUREMENT, 57, 1313-1321.

769 Bhattacharyya, N., Seth, S., Tudu, B., Tamuly, P., Jana, A., Ghosh, D., Bandyopadhyay, R., &

770 Bhuyan, M. (2007). Monitoring of black tea fermentation process using electronic nose.

771 Journal of Food Engineering, 80, 1146-1156.

772 Bhattacharyya, N., Seth, S., Tudu, B., Tamuly, P., Jana, A., Ghosh, D., Bandyopadhyay, R., Bhuyan,
 ACCEPTED MANUSCRIPT

773 M., & Sabhapandit, S. (2007). Detection of optimum fermentation time for black tea

774 manufacturing using electronic nose. Sensors and Actuators B: Chemical, 122, 627-634.

775 Bhattacharyya, N., Tudu, B., & Metla, A. (2008). Incremental PNN Classifier for a Versatile

776 Electronic Nose. In S. Mukhopadhyay, G. Gupta & R. Huang (Eds.), 3rd International

777 Conference on Sensing Technology (pp. 242-247). Natl Cheng Kung Univ, Tainan, TAIWAN:

PT
778 Proceeding of the Third International Conference on Sensing Technology.

779 Bhattacharyya, R., Tudu, B., Das, S. C., Bhattacharyya, N., Bandyopadhyay, R., & Pramanik, P.

RI
780 (2012). Classification of black tea liquor using cyclic voltammetry. Journal of Food

781 Engineering, 109, 120-126.

SC
782 Bhondekar, A. P., Dhiman, M., Sharma, A., Bhakta, A., Ganguli, A., Bari, S. S., Vig, R., Kapur, P.,

783 & Singla, M. L. (2010). A novel iTongue for Indian black tea discrimination. Sensors and

U
784 Actuators B-Chemical, 148, 601-609.
AN
785 Bhondekar, A. P., Kaur, R., Kumar, R., Vig, R., & Kapur, P. (2011). A novel approach using

786 Dynamic Social Impact Theory for optimization of impedance-Tongue (iTongue).

M

787 Chemometrics and Intelligent Laboratory Systems, 109, 65-76.

788 Bhoo Pathy, N., Peeters, P., van Gils, C., Beulens, J. W. J., van der Graaf, Y., Bueno-de-Mesquita, B.,
D

789 Bulgiba, A., & Uiterwaal, C. S. P. M. (2010). Coffee and tea intake and risk of breast cancer.
TE

790 Breast cancer research and treatment, 121, 461-467.

791 Brosnan, T., & Sun, D.-W. (2004). Improving quality inspection of food products by computer
EP

792 vision––a review. Journal of Food Engineering, 61, 3-16.

793 Casalinuovo, I. A., Pierro, D. D., Coletta, M., & Francesco, P. D. (2006). Application of Electronic
C

794 Noses for Disease Diagnosis and Food Spoilage Detection. Sensors, 6, 1428-1439.

795 Cen, H. Y., & He, Y. (2007). Theory and application of near infrared reflectance spectroscopy in
AC

796 determination of food quality. Trends in Food Science & Technology, 18, 72-83.

797 Chen, Q., Liu, A., Zhao, J., & Ouyang, Q. (2013). Classification of tea category using a portable

798 electronic nose based on an odor imaging sensor array. Journal of Pharmaceutical and

799 Biomedical Analysis, 84, 77–83.

800 Chen, Q. S., Guo, Z. M., Zhao, J. W., & Ouyang, Q. (2012). Comparisons of different regressions

801 tools in measurement of antioxidant activity in green tea using near infrared spectroscopy.

802 Journal of Pharmaceutical and Biomedical Analysis, 60, 92-97.

 ACCEPTED MANUSCRIPT

803 Chen, Q. S., Zhao, J. W., & Cai, J. R. (2008). Identification of Tea Varieties Using Computer Vision.

804 Transactions of the ASABE, 51, 623-628.

805 Chen, Q. S., Zhao, J. W., Cai, J. R., & Saritporn, V. (2008). Inspection of tea quality by using

806 multi-sensor information fusion based on NIR spectroscopy and machine vision. Transactions

807 of the Chinese Society of Agricultural Engineering., 24, 5-10.

PT
808 Chen, Q. S., Zhao, J. W., Cai, J. R., & Wang, X. Y. (2006). Study on identification of tea using

809 computer vision based on support vector machine [J]. Chinese Journal of Scientific Instrument,

RI
810 12, 031.

811 Chen, Q. S., Zhao, J. W., Chaitep, S. P., & Guo, Z. M. (2009). Simultaneous analysis of main

SC
812 catechins contents in green tea (Camellia sinensis (L.)) by Fourier transform near infrared

813 reflectance (FT-NIR) spectroscopy. Food Chemistry, 113, 1272-1277.

U
814 Chen, Q. S., Zhao, J. W., Chen, Z., Lin, H., & Zhao, D.-A. (2011). Discrimination of green tea
AN
815 quality using the electronic nose technique and the human panel test, comparison of linear and

816 nonlinear classification tools. Sensors and Actuators B: Chemical, 159, 294-300.
M

817 Chen, Q. S., Zhao, J. W., Fang, C. H., & Wang, D. M. (2007). Feasibility study on identification of

818 green, black and Oolong teas using near-infrared reflectance spectroscopy based on support
D

819 vector machine (SVM). Spectrochimica Acta Part a-Molecular and Biomolecular
TE

820 Spectroscopy, 66, 568-574.

821 Chen, Q. S., Zhao, J. W., Guo, Z. M., & Wang, X. Y. (2010). Determination of caffeine content and
EP

822 main catechins contents in green tea (Camellia sinensis L.) using taste sensor technique and

823 multivariate calibration. Journal of Food Composition and Analysis, 23, 353-358.
C

824 Chen, Q. S., Zhao, J. W., Huang, X. Y., Zhang, H. D., & Liu, M. H. (2006). Simultaneous

825 determination of total polyphenols and caffeine contents of green tea by near-infrared
AC

826 reflectance spectroscopy. Microchemical Journal, 83, 42-47.

827 Chen, Q. S., Zhao, J. W., & Lin, H. (2009). Study on discrimination of Roast green tea (Camellia

828 sinensis L.) according to geographical origin by FT-NIR spectroscopy and supervised pattern

829 recognition. Spectrochimica Acta Part a-Molecular and Biomolecular Spectroscopy, 72,

830 845-850.

831 Chen, Q. S., Zhao, J. W., Liu, M. H., & Cai, J. R. (2008). Nondestructive Identification of Tea

832 (Camellia sinensis L.) Varieties using FT-NIR Spectroscopy and Pattern Recognition. Czech
 ACCEPTED MANUSCRIPT

833 Journal of Food Sciences, 26, 360-367.

834 Chen, Q. S., Zhao, J. W., Liu, M. H., Cai, J. R., & Liu, J. H. (2008). Determination of total

835 polyphenols content in green tea using FT-NIR spectroscopy and different PLS algorithms.

836 Journal of Pharmaceutical and Biomedical Analysis, 46, 568-573.

837 Chen, Q. S., Zhao, J. W., & Vittayapadung, S. (2008). Identification of the green tea grade level

PT
838 using electronic tongue and pattern recognition. Food Research International, 41, 500-504.

839 Chen, Q. S., Zhao, J. W., Zhang, H. D., & Fang, M. (2005). Identification of tea color by using

RI
840 computer vision [J]. Journal of Jiangsu University (National Science Edition), 6, 001.

841 Chen, Q. S., Zhao, J. W., Zhang, H. D., Liu, M. H., & Fang, M. (2005). Qualitative identification of

SC
842 tea by near infrared spectroscopy based on soft independent modelling of class analogy pattern

843 recognition. Journal of near Infrared Spectroscopy, 13, 327-332.

U
844 Chen, Q. S., Zhao, J. W., Zhang, H. D., & Wang, X. Y. (2006). Feasibility study on qualitative and
AN
845 quantitative analysis in tea by near infrared spectroscopy with multivariate calibration.

846 Analytica Chimica Acta, 572, 77-84.

M

847 Chen, X. J., Wu, D., & He, Y. (2008). Study on Discrimination of Tea Based on Color of

848 Multispectral Image. Spectroscopy and Spectral Analysis, 28, 2527-2530.

D

849 Ciosek, P., & Wroblewski, W. (2007). Sensor arrays for liquid sensing - electronic tongue systems.
TE

850 Analyst, 132, 963-978.

851 de la Guardia, M. (2010). Green analytical chemistry. Trac-Trends in Analytical Chemistry, 29,
EP

852 577-577.

853 Deka, A., & Vita, J. A. (2011). Tea and cardiovascular disease. Pharmacological Research, 64,
C

854 136-145.

855 Del Rio, D., Stewart, A. J., Mullen, W., Burns, J., Lean, M. E., Brighenti, F., & Crozier, A. (2004).
AC

856 HPLC-MSn analysis of phenolic compounds and purine alkaloids in green and black tea.

857 Journal of agricultural and food chemistry, 52, 2807-2815.

858 Dominguez-Perles, R., Moreno, D. A., Carvajal, M., & Garcia-Viguera, C. (2011). Composition and

859 antioxidant capacity of a novel beverage produced with green tea and minimally-processed

860 byproducts of broccoli. Innovative Food Science & Emerging Technologies.

861 Dutta, R., Hines, E. L., Gardner, J. W., Kashwan, K. R., & Bhuyan, M. (2003). Tea quality

862 prediction using a tin oxide-based electronic nose: an artificial intelligence approach. Sensors
 ACCEPTED MANUSCRIPT

863 and Actuators B: Chemical, 94, 228-237.

864 Escuder-Gilabert, L., & Peris, M. (2010). Review: Highlights in recent applications of electronic

865 tongues in food analysis. Analytica Chimica Acta, 665, 15-25.

866 Evtugyn, G. A., Shamagsumova, R. V., Stoikova, E. E., Sitdikov, R. R., Stoikov, II, Budnikov, H. C.,

867 Ivanov, A. N., & Antipin, I. S. (2011). Potentiometric Sensors Based on Polyaniline and

PT
868 Thiacalixarenes for Green Tea Discrimination. Electroanalysis, 23, 1081-1088.

869 Gill, G. S., Kumar, A., & Agarwal, R. (2011). Monitoring and grading of tea by computer vision – A

RI
870 review. Journal of Food Engineering, 106, 13-19.

871 Gowen, A., Odonnell, C., Cullen, P., Downey, G., & Frias, J. (2007). Hyperspectral imaging – an

SC
872 emerging process analytical tool for food quality and safety control. Trends in Food Science &

873 Technology, 18, 590-598.

U
874 Graham, H. N. (1992). Green tea composition, consumption, and polyphenol chemistry. Preventive
AN
875 medicine, 21, 334-350.

876 Guan, X., Chen, X., & Jiang, J. (2011). Integration of modified uninformative variable elimination
M

877 and successive projections algorithm for determination harvest time of laver by using visible

878 and near infrared spectra. African Journal of Agricultural Research, 6, 5987-5991.
D

879 Guo, Z. M., Chen, Q. S., Chen, L. P., Huang, W. Q., Zhang, C., & Zhao, C. J. (2011). Optimization
TE

880 of Informative Spectral Variables for the Quantification of EGCG in Green Tea Using Fourier

881 Transform Near-Infrared (FT-NIR) Spectroscopy and Multivariate Calibration. Applied

EP

882 Spectroscopy, 65, 1062-1067.

883 Hall, M. N., Robertson, A., & Scotter, C. N. G. (1988). Near-infrared reflectance prediction of
C

884 quality, theaflavin content and moisture content of black tea. Food Chemistry, 27, 61-65.

885 He, W., Hu, X. S., Zhao, L., Liao, X. J., Zhang, Y., Zhang, M. W., & Wu, J. H. (2009). Evaluation of
AC

886 Chinese tea by the electronic tongue: Correlation with sensory properties and classification

887 according to geographical origin and grade level. Food Research International, 42, 1462-1467.

888 He, Y., Li, X. L., & Deng, X. F. (2007). Discrimination of varieties of tea using near infrared

889 spectroscopy by principal component analysis and BP model. Journal of Food Engineering, 79,

890 1238-1242.

891 Hsu, S. (2005). Green tea and the skin. Journal of the American Academy of Dermatology, 52,

892 1049-1059.
 ACCEPTED MANUSCRIPT

893 Huang, H. B., Yu, H. Y., Xu, H. R., & Ying, Y. B. (2008). Near infrared spectroscopy for on/in-line

894 monitoring of quality in foods and beverages: A review. Journal of Food Engineering, 87,

895 303-313.

896 Iiyama, S., Ezaki, S., Toko, K., Matsuno, T., & Yamafuji, K. (1995). Study of astringency and

897 pungency with multichannel taste sensor made of lipid membranes. Sensors and Actuators B:

PT
898 Chemical, 24, 75-79.

899 Ikeda, T., Kanaya, S., Yonetani, T., Kobayashi, A., & Fukusaki, E. (2007). Prediction of Japanese

RI
900 green tea ranking by Fourier transform near-infrared reflectance spectroscopy. Journal of

901 agricultural and food chemistry, 55, 9908-9912.

SC
902 Ilia, N., Guang, L., & Alexander, G. (2010). Application of Conformal Predictors to Tea

903 Classification Based on Electronic Nose. In 6th IFIP Conference on Artificial Intelligence

U
904 Applications and Innovations (Vol. 339, pp. 303-310). Larnaca, CYPRUS: Artificial
AN
905 Intelligence Applications and Innovations.

906 Isaksson, T., & Næs, T. (1988). The Effect of Multiplicative Scatter Correction (MSC) and Linearity
M

907 Improvement in NIR Spectroscopy. Applied Spectroscopy, 42, 1273-1284.

908 Ivarsson, P., Holmin, S., Hojer, N. E., Krantz-Rulcker, C., & Winquist, F. (2001). Discrimination of
D

909 tea by means of a voltammetric electronic tongue and different applied waveforms. Sensors
TE

910 and Actuators B-Chemical, 76, 449-454.

911 James, D., Scott, S. M., Ali, Z., & O’hare, W. T. (2005). Chemical sensors for electronic nose
EP

912 systems. Microchimica Acta, 149, 1-17.

913 Jiang, F., Qiao, X., Zheng, H., & Yang, Q. (2011). Grade discrimination of machine-fried Longjing
C

914 tea based on hyperspectral technology. Transactions of the Chinese Society of Agricultural

915 Engineering, 27, 343-348.

AC

916 Kovács, Z., Dalmadi, I., Lukács, L., Sipos, L., Szántai-Kőhegyi, K., Kókai, Z., & Fekete, A. (2010).

917 Geographical origin identification of pure Sri Lanka tea infusions with electronic nose,

918 electronic tongue and sensory profile analysis. Journal of Chemometrics, 24, 121-130.

919 Laddi, A., Prakash, N. R., Sharma, S., Mondal, H. S., Kumar, A., & Kapur, P. (2012). Significant

920 physical attributes affecting quality of Indian black (CTC) tea. Journal of Food Engineering,

921 113, 69-78.

922 Laddi, A., Sharma, S., Kumar, A., & Kapur, P. (2013). Classification of tea grains based upon image
 ACCEPTED MANUSCRIPT

923 texture feature analysis under different illumination conditions. Journal of Food Engineering,

924 115, 226-231.

925 Li, P., Wang, J., Cao, N., & Han, D. (2009). Selection of variables for MLR in Vis/NIR spectroscopy

926 based on BiPLS combined with GA]. Spectroscopy and Spectral Analysis, 29, 2637.

927 Li, X. L., & He, Y. (2009). Classification of Tea Grades by Multi-spectral Images and Combined

PT
928 Features. Transactions of the Chinese Society for Agricultural Machinery, 40, 113-118.

929 Li, X. L., He, Y., & Qiu, Z. J. (2007). Application PCA-ANN method to fast discrimination of tea

RI
930 varieties using visible/near infrared spectroscopy. Spectroscopy and Spectral Analysis, 27,

931 279-282.

SC
932 Li, X. L., He, Y., & Qiu, Z. J. (2008). Textural feature extraction and optimization in wavelet

933 sub-bands for discrimination of green tea brands. In 7th International Conference on Machine

U
934 Learning and Cybernetics (pp. 1461-1466). Kunming, PEOPLES R CHINA.
AN
935 Li, X. L., He, Y., Qiu, Z. J., & Bao, Y. D. (2008). Tea category classification using morphological

936 characteristics and support vector machines. In 28th International Congress on High-Speed
M

937 Imaging and Photonics (Vol. 7126). Canberra, AUSTRALIA.

938 Li, X. L., He, Y., Qiu, Z. J., & Wu, D. (2007). Texture discrimination of green tea categories based
D

939 on least squares support vector machine (LSSVM) classifier. In International Conference on
TE

940 Photoelectronic Detection and Imaging (Vol. 6625, pp. 662516-662516-662512). Beijing,

941 PEOPLES R CHINA.

EP

942 Liu, S. L., Tsai, Y. S., & Ou, A. S. M. (2010). Classifying the Variety, Production Area and Season of

943 Taiwan Partially Fermented Tea by Near Infrared Spectroscopy. Journal of Food and Drug
C

944 Analysis, 18, 34-43.

945 Lorber, A. (1986). Error propagation and figures of merit for quantification by solving matrix
AC

946 equations. Analytical Chemistry, 58, 1167-1172.

947 Lu, R. F., & Chen, Y. R. (1998). Hyperspectral imaging for safety inspection of food and agricultural

948 products. In SPIE Conference on Pathogen Detection and Remediation for Safe Eating.

949 Boston.

950 Lu, Y. J., Chen, H. C., Lü, J., & Chen, X. D. (2005). Near infrared determination of the content of

951 caffeine in tea polyphenol. Spectroscopy and Spectral Analysis, 25, 1243.

952 Luo, Y. F., Guo, Z. F., Zhu, Z. Y., Wang, C. P., Jiang, H. Y., & Han, B. Y. (2005). Studies on ANN
 ACCEPTED MANUSCRIPT

953 models of determination of tea polyphenol and amylose in tea by near-infrared spectroscopy.

954 Spectroscopy and Spectral Analysis, 25, 1230-1233.

955 Lvova, L., Legin, A., Vlasov, Y., Cha, G. S., & Nam, H. (2003). Multicomponent analysis of Korean

956 green tea by means of disposable all-solid-state potentiometric electronic tongue microsystem.

957 Sensors and Actuators B-Chemical, 95, 391-399.

PT
958 Mandel, S. A., Avramovich-Tirosh, Y., Reznichenko, L., Zheng, H., Weinreb, O., Amit, T., &

959 Youdim, M. (2005). Multifunctional activities of green tea catechins in neuroprotection.

RI
960 Neurosignals, 14, 46-60.

961 Manley, M., Joubert, E., & Botha, M. (2006). Quantification of the major phenolic compounds,

SC
962 soluble solid content and total antioxidant activity of green rooibos (Aspalathus linearis) by

963 means of near infrared spectroscopy. Journal of near Infrared Spectroscopy, 14, 213.

U
964 Mukhtar, H., & Ahmad, N. (2000). Tea polyphenols: prevention of cancer and optimizing health.
AN
965 The American journal of clinical nutrition, 71, 1698S-1702S.

966 Nagle, H. T., Gutierrez-Osuna, R., & Schiffman, S. S. (1998). The how and why of electronic noses.
M

967 Spectrum, IEEE, 35, 22-31.

968 Nazifah, A. F., Shakaff, M., Hamid, A., Adom, P., Noor, M., Ahmad, P., Ammar, Z., & MA, M.
D

969 (2011). Development of human sensory mimicking system. Sensor Letters, 9, 423-427.
TE

970 Nieh, C. H., Hsieh, B. C., Chen, P. C., Hsiao, H. Y., Cheng, T. J., & Chen, R. L. C. (2009).

971 Potentiometric flow-injection estimation of tea fermentation degree. Sensors and Actuators
EP

972 B-Chemical, 136, 541-545.

973 Niu, W. (2012). A chemiluminescence sensor array based on nanomaterials for discrimination of
C

974 teas. Luminescence.

975 Olsson, J., Winquist, F., & Lundström, I. (2006). A self polishing electronic tongue. Sensors and
AC

976 Actuators B: Chemical, 118, 461-465.

977 Ono, D., Bamba, T., Oku, Y., Yonetani, T., & Fukusaki, E. (2011). Application of Fourier transform

978 near-infrared spectroscopy to optimization of green tea steaming process conditions. Journal

979 of Bioscience and Bioengineering, 112, 247-251.

980 Palit, M., Bhattacharyya, N., Sarkar, S., Dutta, A., Dutta, P. K., Tudu, B., Bandyopadhyay, R., &

981 Ieee. (2008). Virtual Instrumentation Based Voltammetric Electronic Tongue for Classification

982 of Black Tea.

 ACCEPTED MANUSCRIPT

983 Palit, M., Tudu, B., Bhattacharyya, N., Dutta, A., Dutta, P. K., Jana, A., Bandyopadhyay, R., &

984 Chatterjee, A. (2010). Comparison of multivariate preprocessing techniques as applied to

985 electronic tongue based pattern classification for black tea. Analytica Chimica Acta, 675, 8-15.

986 Palit, M., Tudu, B., Dutta, P. K., Dutta, A., Jana, A., Roy, J. K., Bhattacharyya, N., Bandyopadhyay,

987 R., & Chatterjee, A. (2010). Classification of Black Tea Taste and Correlation With Tea

PT
988 Taster's Mark Using Voltammetric Electronic Tongue. Ieee Transactions on Instrumentation

989 and Measurement, 59, 2230-2239.

RI
990 Patel, K. K., Kar, A., Jha, S. N., & Khan, M. A. (2011). Machine vision system: a tool for quality

991 inspection of food and agricultural products. Journal of Food Science and Technology, 49,

SC
992 123-141.

993 Persaud, K., & Dodd, G. (1982). Analysis of discrimination mechanisms in the mammalian

U
994 olfactory system using a model nose. Nature, 299, 352-355.
AN
995 Phaisangittisagul, E., & Nagle, H. T. (2011). Predicting odor mixture's responses on machine

996 olfaction sensors. Sensors and Actuators B: Chemical, 155, 473-482.

M

997 Pioggia, G., Di Francesco, F., Marchetti, A., Ferro, M., & Ahluwalia, A. (2007). A composite sensor

998 array impedentiometric electronic tongue Part I. Characterization. Biosensors &

D

999 Bioelectronics, 22, 2618-2623.

TE

1000 Pongsuwan, W., Bamba, T., Yonetani, T., Kobayashi, A., & Fukusaki, E. (2008). Quality prediction

1001 of Japanese green tea using pyrolyzer coupled GC/MS based metabolic fingerprinting. Journal
EP

1002 of agricultural and food chemistry, 56, 744-750.

1003 Riul, A., Malmegrim, R. R., Fonseca, F. J., & Mattoso, L. H. C. (2003). An artificial taste sensor
C

1004 based on conducting polymers. Biosensors & Bioelectronics, 18, 1365-1369.

1005 Rohman, A., & Man, Y. (2010). Fourier transform infrared (FTIR) spectroscopy for analysis of extra
AC

1006 virgin olive oil adulterated with palm oil. Food Research International, 43, 886-892.

1007 Roy, P., Tsai, T. C., Liang, C. T., Wu, R. J., & Chavali, M. (2011). Application of Impedance

1008 Measurement Technology in Distinguishing Different Tea Samples with Ppy/SWCNT

1009 Composite Sensing Material. Journal of the Chinese Chemical Society, 58, 714-722.

1010 Sankar, C. S., Tudu, B., Bhattacharyya, N., & Bandyopadhyay, R. (2008). Portable Electronic Nose

1011 System for Aroma Classification of Black Tea. IEEE Region 10 Colloquiumand the Third

1012 ICIIS, 1, 387-391.

 ACCEPTED MANUSCRIPT

1013 Savitzky, A., & Golay, M. J. E. (1964). Smoothing and Differentiation of Data by Simplified Least

1014 Squares Procedures. Analytical Chemistry, 36, 1627-1639.

1015 Schulz, H., Engelhardt, U. H., Wegent, A., Drews, H. H., & Lapczynski, S. (1999). Application of

1016 Near-Infrared Reflectance Spectroscopy to the Simultaneous Prediction of Alkaloids and

1017 Phenolic Substances in Green Tea Leaves. Journal of agricultural and food chemistry, 47,

PT
1018 5064-5067.

1019 Schwender, H., Zucknick, M., Ickstadt, K., & Bolt, H. M. (2004). A pilot study on the application of

RI
1020 statistical classification procedures to molecular epidemiological data. Toxicology letters, 151,

1021 291-299.

SC
1022 Setiawan, V. W., Zhang, Z. F., Yu, G. P., Lu, Q. Y., Lu, M. L., Wang, M. R., & Guo, C. H. (2001).

1023 Protective effect of green tea on the risks of chronic gastritis and stomach cancer. International

U
1024 Journal of Cancer, 92, 600-604.
AN
1025 Sinija, V. R., & Mishra, H. N. (2009). FT-NIR spectroscopy for caffeine estimation in instant green

1026 tea powder and granules. Lwt-Food Science and Technology, 42, 998-1002.
M

1027 Soylak, M., Tuzen, M., Souza, A. S., Korn, M. d. G. A., & Ferreira, S. L. C. (2007). Optimization of

1028 microwave assisted digestion procedure for the determination of zinc, copper and nickel in tea
D

1029 samples employing flame atomic absorption spectrometry. Journal of Hazardous materials,
TE

1030 149, 264-268.

1031 Sun, X., Hao, Y., Cai, L., & Liu, Y. (2011). Selection of visible-NIR variables based on extraction
EP

1032 and successive projections algorithm. Spectroscopy and Spectral Analysis, 31, 2399.

1033 Tan, S. M., Luo, R. M., Zhou, Y. P., Xu, H., Song, D. D., Ze, T., Yang, T. M., & Nie, Y. (2012).
C

1034 Boosting partial least-squares discriminant analysis with application to near infrared

1035 spectroscopic tea variety discrimination. Journal of Chemometrics, 26, 34-39.

AC

1036 Teh, S., Zheng, W., Ho, K., Teh, M., Yeoh, K., & Huang, Z. (2008). Diagnostic potential of

1037 near-infrared Raman spectroscopy in the stomach: differentiating dysplasia from normal tissue.

1038 British journal of cancer, 98, 457-465.

1039 Tian, G. Y., Yuan, H. F., Liu, H. Y., & Lu, W. Z. (2003). The application of wavelet transform in near

1040 infrared spectroscopy. Spectroscopy and Spectral Analysis, 23, 1111-1114.

1041 Tian, S. Y., Deng, S. P., & Chen, Z. X. (2007). Multifrequency large amplitude pulse voltammetry:

1042 A novel electrochemical method for electronic tongue. Sensors and Actuators B-Chemical,
 ACCEPTED MANUSCRIPT

1043 123, 1049-1056.

1044 Trevisanato, S. I., & Kim, Y. I. (2000). Tea and health. Nutrition Reviews, 58, 1-10.

1045 Tudu, B., Bhattacharyya, N., & Bikram, K. (2008). Comparison of Multivariate Normalization

1046 Techniques as Applied to Electronic Nose Based Pattern Classification for Black Tea.

1047 PROCEEDINGS OF THE THIRD INTERNATIONAL CONFERENCE ON SENSING

PT
1048 TECHNOLOGY, 254-258.

1049 Tudu, B., Jana, A., Metla, A., Ghosh, D., Bhattacharyya, N., & Bandyopadhyay, R. (2009).

RI
1050 Electronic nose for black tea quality evaluation by an incremental RBF network. Sensors and

1051 Actuators B: Chemical, 138, 90-95.

SC
1052 Tudu, B., Metla, A., & Das, B. (2009). Towards Versatile Electronic Nose Pattern Classifier for

1053 Black Tea Quality Evaluation An Incremental Fuzzy Approach. IEEE Transactions on

U
1054 Instrumentation and Measurement, 58, 3069-3078.
AN
1055 Vlasov, Y., Legin, A., Rudnitskaya, A., Di Natale, C., & D'Amico, A. (2005). Nonspecific sensor

1056 arrays ("electronic tongue") for chemical analysis of liquids (IUPAC Technical Report). Pure
M

1057 and Applied Chemistry, 77, 1965-1983.

1058 Wang, H., Helliwell, K., & You, X. (2000). Isocratic elution system for the determination of
D

1059 catechins, caffeine and gallic acid in green tea using HPLC. Food Chemistry, 68, 115-121.
TE

1060 Wang, J., Zeng, X. Y., & Du, S. P. (2010). Identification and Grading of Tea Using Computer Vision.

1061 Applied Engineering in Agriculture., 26, 639-645.

EP

1062 Weisburger, J., & Chung, F. L. (2002). Mechanisms of chronic disease causation by nutritional

1063 factors and tobacco products and their prevention by tea polyphenols. Food and Chemical
C

1064 Toxicology, 40, 1145-1154.

1065 Westerhuis, J. A., de Jong, S., & Smilde, A. K. (2001). Direct orthogonal signal correction.
AC

1066 Chemometrics and Intelligent Laboratory Systems, 56, 13-25.

1067 Weyer, L. (1985). Near-infrared spectroscopy of organic substances. Applied Spectroscopy Reviews,

1068 21, 1-43.

1069 Winquist, F. (2008). Voltammetric electronic tongues - basic principles and applications.

1070 Microchimica Acta, 163, 3-10.

1071 Winquist, F., Wide, P., & Lundström, I. (1997). An electronic tongue based on voltammetry.

1072 Analytica Chimica Acta, 357, 21-31.

 ACCEPTED MANUSCRIPT

1073 Wolda, S., Anttia, H., Lindgrenb, F., & Öhmanc, J. (1998). Orthogonal signal correction of

1074 near-infrared spectra. Chemometrics and Intelligent Laboratory Systems, 44, 175-185.

1075 Workman Jr, J. J. (1996). Interpretive spectroscopy for near infrared. Applied Spectroscopy Reviews,

1076 31, 251-320.

1077 Wu, D., Chen, X. J., & He, Y. (2009). Application of Multispectral Image Texture to Discriminating

PT
1078 Tea Categories Based on DCT and LS-SVM. Spectroscopy and Spectral Analysis, 29,

1079 1382-1385.

RI
1080 Wu, R. M., Zhao, J. W., Chen, Q. S., & Huang, X. Y. (2011). Determination of Taste Quality of

1081 Green Tea Using FT-NIR Spectroscopy and Variable Selection Methods. Spectroscopy and

SC
1082 Spectral Analysis, 31, 1782-1785.

1083 Xiao, H., & Wang, J. (2009). Discrimination of Xihulongjing tea grade using an electronic tongue.

U
1084 African Journal of Biotechnology, 8, 6985-6992.
AN
1085 Yan, S. H. (2005). Evaluation of the composition and sensory properties of tea using near infrared

1086 spectroscopy and principal component analysis. Journal of near Infrared Spectroscopy, 13,
M

1087 313-325.

1088 Yang, Fu, J., Lou, Z. G., Wang, L. Y., Li, G., & Freeman, W. J. (2006). Tea classification based on
D

1089 artificial olfaction using bionic olfactory neural network. Advances in Neural Networks., 3972,
TE

1090 343-348.

1091 Yang, Z. Y., Dong, F., Shimizu, K., Kinoshita, T., Kanamori, M., Morita, A., & Watanabe, N. (2009).
EP

1092 Identification of coumarin-enriched Japanese green teas and their particular flavor using

1093 electronic nose. Journal of Food Engineering, 92, 312-316.

C

1094 Yu, H. C., & Wang, J. (2007). Discrimination of LongJing green-tea grade by electronic nose.

1095 Sensors and Actuators B: Chemical, 122, 134-140.

AC

1096 Yu, H. C., Wang, J., Xiao, H., & Liu, M. (2009). Quality grade identification of green tea using the

1097 eigenvalues of PCA based on the E-nose signals. Sensors and Actuators B: Chemical, 140,

1098 378-382.

1099 Yu, H. C., Wang, J., Yao, C., Zhang, H. M., & Yu, Y. (2008). Quality grade identification of green

1100 tea using E-nose by CA and ANN. LWT - Food Science and Technology, 41, 1268-1273.

1101 Yu, H. C., Wang, J., Zhang, H. M., Yu, Y., & Yao, C. (2008). Identification of green tea grade using

1102 different feature of response signal from E-nose sensors. Sensors and Actuators B: Chemical,
 ACCEPTED MANUSCRIPT

1103 128, 455-461.

1104 Yu, H. C., Wang, Y. W., & Wang, J. (2009). Identification of Tea Storage Times by Linear

1105 Discrimination Analysis and Back-Propagation Neural Network Techniques Based on the

1106 Eigenvalues of Principal Components Analysis of E-Nose Sensor Signals. Sensors, 9,

1107 8073-8082.

PT
1108 Zakaria, A., Shakaff, A. Y. M., Adom, A. H., Ahmad, M. N., Masnan, M. J., Aziz, A. H. A., Fikri, N.

1109 A., Abdullah, A. H., & Kamarudin, L. M. (2010). Improved classification of orthosiphon

RI
1110 stamineus by data fusion of electronic nose and tongue sensors. Sensors, 10, 8782-8796.

1111 Zhang, M. H., Luypaert, J., Pierna, J. A. F., Xu, Q. S., & Massart, D. L. (2004). Determination of

SC
1112 total antioxidant capacity in green tea by near-infrared spectroscopy and multivariate

1113 calibration. Talanta, 62, 25-35.

U
1114 Zhao, J. W., Chen, Q. S., Cai, J. R., & Ouyang, Q. (2009). Automated tea quality classification by
AN
1115 hyperspectral imaging. Applied Optics., 48, 3557-3564.

1116 Zhao, J. W., Chen, Q. S., Huang, X. Y., & Fang, C. H. (2006). Qualitative identification of tea
M

1117 categories by near infrared spectroscopy and support vector machine. Journal of

1118 Pharmaceutical and Biomedical Analysis, 41, 1198-1204.

D

1119 Zhao, J. W., Wang, K. L., Ouyang, Q., & Chen, Q. S. (2011). Measurement of Chlorophyll Content
TE

1120 and Distribution in Tea Plant's Leaf Using Hyperspectral Imaging Technique. Spectroscopy

1121 and Spectral Analysis, 31, 512-515.

EP

1122 Zhou, J., Cheng, H., He, W., Wang, L. Y., Liu, X., & Lu, W. Y. (2009). Short communication:

1123 Identification of geographical indication tea with Fisher's discriminant classification and
C

1124 principal components analysis. Journal of near Infrared Spectroscopy, 17, 159-164.

1125 Zhou, Y., Wang, D., Wan, X., Zhang, L., Du, X., & Hu, W. (2009). Effect of tea polysaccharide
AC

1126 addition on the properties of bread made from two flours. Journal of Food Processing and

1127 Preservation, 33, 798-813.

1128 Zou, X. B., Zhao, J. W., Povey, M. J. W., Holmes, M., & Mao, H. P. (2010). Variables selection

1129 methods in near-infrared spectroscopy. Analytica Chimica Acta, 667, 14-32.

1130 Zuo, Y., Chen, H., & Deng, Y. (2002). Simultaneous determination of catechins, caffeine and gallic

1131 acids in green, Oolong, black and pu-erh teas using HPLC with a photodiode array detector.

1132 Talanta, 57, 307-316.

 ACCEPTED MANUSCRIPT

1133 Figures captions

1134 Fig. 1. Scatters plots showing the correlation between reference measurement and NIR

1135 prediction in the calibration set (＊) and the prediction set (o) for four typical catechins

1136 contents.

PT
1137 Fig. 2. Sketch of E-tongue system used for classification of different grades of tea.

RI
SC
1138 Fig. 3. Sketch of study on application of a developed E-nose system in discrimination

1139 of tea grades.

U
AN
1140 Fig. 4. Sketch of application of an odor imaging sensor array in tea classification.

Fig. 5. Three-dimensional hypercube for tea sample.

M

1141
D

1142 Fig. 6. Flowchart of data fusion for inspection of tea quality by combining two green
TE

1143 analytical technologies of NIR spectroscopy and computer imaging.

C EP
AC
 ACCEPTED MANUSCRIPT

Tables 1-5

Table 1 Algorithms in multivariate calibration of NIR spectral data

Methods Algorithms Functions References

Spectral Smoothing Smoothing is often used to eliminate noises; generally (Savitzky & Golay,

PT
preprocessing used combined with other pre-processing methods 1964)

methods Derivative Account for the variation in baseline shift and

(Barnes, et al.,

RI
curvilinearity in the spectra, and to remove background
1989)
and increase spectral resolution

SC
SNV Remove the multiplicative interferences of scatter and (Barnes, et al.,

particle size in NIR spectra 1989)

U
MSC Correct both multiplicative and additive scatter effects,
(Isaksson & Næs,
AN
and remove slope variations on individual spectrum
1988)
basis
M

WT Separating the feature spectra of original signals (G. Y. Tian, et al.,

2003)
D

OSC Remove from the spectral matrix (X) only the part that
(Wolda, et al.,
TE

definitely is unrelated to Y (i.e. mathematically

1998)
orthogonal to Y, or as close to orthogonal as possible).
EP

DOSC Calculate components that are orthogonal to Y and (Westerhuis, et al.,

describe the largest variation in X 2001)

C

NAS Estimate selectivities (ratio of signal available for

(Lorber, 1986)
quantitation to the total measured signal)
AC

Variables SPA Extract variables for developing partial least squares (Sun, Hao, Cai, &

selection (PLS) models from the pretreatment spectra Liu, 2011)

methods UVE Used to eliminate uninformative variables and noise in (Guan, Chen, &

the spectral data Jiang, 2011)

PCA Compress the dimensions of the original variables (NIR (Chen, Zhao, &

spectra data) to new PCs Lin, 2009)

 ACCEPTED MANUSCRIPT

iPLS Select the most efficient spectral variables

GAPLS
(Guo, et al., 2011)
SiPLS

BiPLS/FiPLS

Linear LDA Maximizes the variance between categories and

PT
methods minimizes the variance within categories (Teh, et al., 2008)

KNN An unknown observation is predicted by finding the K

RI
(Schwender,
observations in training set that are closest in distance to
Zucknick, Ickstadt,

SC
this new observation and by choosing the class to which
& Bolt, 2004)
most of the K observations belong

U
MLR More dependent on a good choice of spectral variables,
(Li, Wang, Cao, &
and it is simple and easy to interpret
AN
Han, 2009)

PCR Compress the dimensions of the original independent

M

variables by constructing the relationship between the (Isaksson & Næs,

original independent variables and new reduced 1988)

D

dimension independent variables.

TE

PLS Relate independent variables such as the FT-NIR spectral (Ono, Bamba,

data with dependent variables such as the steaming Oku, Yonetani, &
EP

parameter sets Fukusaki, 2011)

Nonlinear SVM Map the complex and nonlinear data into a higher
(Chen, Guo, Zhao,
C

methods dimensional feature space, where the nonlinear problem

& Ouyang, 2012)
AC

could be solved by a linear tool

ANN Use several simple units to construct a network system (Luo, et al., 2005)
 ACCEPTED MANUSCRIPT

Table 2 Summary of NIR application in tea analysis

Type of study Properties of study Spectral Model calibration References

preprocessing

Qualitative Tea categories SNV; derivative; PCA; SVM; (Chen, Zhao, Fang, & Wang,

analysis S-G smooth BP-ANN 2007; Zhao, Chen, Huang, &

PT
Fang, 2006)

RI
Geographical SNV; MSC PCA; LDA; KNN; (Chen, Zhao, & Lin, 2009; Liu,

indication of tea ANN; SVM; FDA; Tsai, & Ou, 2010; Zhou, et al.,

SC
PLS 2009)

Tea varieties SNV; MC; MSC; PCA; PLS; SIMCA; (Chen, Zhao, Liu, & Cai, 2008;

U
WT AN LDA; ANN; Chen, Zhao, Zhang, Liu, &

PLS-DA; BPLS-DA; Fang, 2005; Chen, Zhao,

BP-ANN Zhang, & Wang, 2006; He, Li,

& Deng, 2007; Li, He, & Qiu,

M

2007; Liu, et al., 2010; Tan, et

D

al., 2012)

Tea leaves of Derivative; PCA (Schulz, Engelhardt, Wegent,

TE

different ages WMCS Drews, & Lapczynski, 1999)

Quantitative Main catechins SNV; MSC iPLS; SiPLS; (Chen, Zhao, Chaitep, et al.,
EP

analysis contents SiPLS-GA; PLS 2009; Guo, et al., 2011)

C

Alkaloids and Derivative; PLS (Schulz, et al., 1999)

phenolic WMCS
AC

substances intea

leaves

Caffeine and total SNV; Derivative; PLS; iPLS; SiPLS (Chen, Zhao, Huang, Zhang, &

polyphenols S-G Liu, 2006; Chen, Zhao, Liu,

content smooth;MSC Cai, & Liu, 2008; Chen, J. W.

Zhao, H. D. Zhang, et al.,

 ACCEPTED MANUSCRIPT

2006; Y. J. Lu, Chen, Lü, &

Chen, 2005; Sinija & Mishra,

2009)

Japanese green tea Derivative; SNV PCA; PLS (Ikeda, Kanaya, Yonetani,

ranking Kobayashi, & Fukusaki, 2007)

PT
Tea plyphenol and No spectral ANN (Luo, et al., 2005)

amylase preprocessing

RI
Composition and No spectral PCA; MLR (Yan, 2005)

sensory properties preprocessing

SC
Major phenolic Offset correction; PCR; PCA; (Manley, Joubert, & Botha,

compounds, detrend; MSC; Sammon’s mapping; 2006; Zhang, et al., 2004;

U
soluble solid SNV; derivative; MD; XRSD; RHM; Chen, et al., 2012)
AN
content and total MIN/MAX MVT;PLS;

antioxidant BP-ANN; SVMR

M

activity

Taste quality SNV SiPLS-GA (Wu, Zhao, Chen, & Huang,

D

2011)
TE

Green tea PCA PLS (Ono, et al., 2011)

steaming process
EP

conditions
C
AC
 ACCEPTED MANUSCRIPT

Table 3 Summary of E-tongue application in tea analysis

Type of
Properties of study Type of sensors Data processing algorithm References
study

Qualitative Tea varieties LAPV, SAPV PCA (Ivarsson, et al., 2001)

analysis and staircase

PT
voltammetry

RI
MLAPV PCA (Tian, Deng, & Chen,

2007)

SC
Tea categories Disposable PCA; SIMCA (Lvova, Legin, Vlasov,

all-solid-state Cha, & Nam, 2003)

U
potentiometry
AN
Impedance PCA (Roy, Tsai, Liang, Wu, &

Chavali, 2011)

Green tea brands Potentiometric PCA; LDA (Evtugyn, et al., 2011)

M

E-tongue
D

Geographical origins Impedance PCA; SITO; GA; bPSO (Bhondekar, et al., 2010;

of tea Bhondekar, et al., 2011; W.

TE

He, et al., 2009)

Tea grades Cyclic Auto scale; Baseline (Bhattacharyya, et al.,

EP

voltammetry subtraction; Range scale; 2012)

PCA; LDA; ANN

C

Voltammetric Range scale; Background (Palit, et al., 2008; Palit,

AC

subtraction; Relative Tudu, Bhattacharyya, et al.,

scale; Baseline 2010; Palit, Tudu, Dutta, et

subtraction; SNV; Auto al., 2010)

scale; PCA; DWT; WT;

LDA; ANN (BP-MLP,

RBF)
 ACCEPTED MANUSCRIPT

Potentiometric PCA; CDA; KNN; ANN; (Chen, Zhao, &

BPNN Vittayapadung, 2008; He,

et al., 2009; Xiao & Wang,

2009)

Tea fermentation Potentiometric LR (Nieh, et al., 2009)

PT
degree flow-injection

(FI) system

RI
Quantitative Caffeine content and Taste sensor PLS; PCA-ANN (Chen, Zhao, Guo, &

analysis main catechins technique Wang, 2010)

SC
contents

Caffeine, catechines, Disposable PCR; PLS (Lvova, et al., 2003)

U
sugar and amino acid all-solid-state
AN
L-arginine content potentiometry
M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT

Table 4 Summary of E-nose application in tea analysis

Type of Properties of Data processing

Type of sensors References
study study algorithm

Qualitative Tea storage times MOS PCA; LDA; BPNN (Yu, Wang, & Wang, 2009)

analysis

PT
Tea grades MOS PCA; KNN; ANN; (Bhattacharya, Tudu, Jana,

SVM; LDA; BPNN; Ghosh, Bandhopadhyaya, &

RI
PNN; CA; Saha, 2008; Bhattacharyya,

incremental PNN Tudu, & Metla, 2008; Chen, et

SC
classifier; al., 2011; Tudu, Metla, & Das,

Incremental Fuzzy 2009; Yu & Wang, 2007; Yu,

U
Classifier Wang, Xiao, & Liu, 2009; Yu,
AN
Wang, Yao, Zhang, & Yu,

2008; Yu, Wang, Zhang, Yu, &

M

Yao, 2008)

Tea types MOS SVM; KIII; BP (Ilia, et al., 2010)

D

networks; conformal
TE

predictor

Tea varieties MOS PCA; CA; AVE; (Bhattacharyya,

EP

FF-MLP; BP-MLP; Bandyopadhyay, & Bhuyan,

ANN; KIII model 2008; Niu, 2012; Sankar,

C

Tudu, Bhattacharyya, &

Bandyopadhyay, 2008; Yang,

AC

et al., 2006; Yang, et al., 2009)

Chemiluminescence LDA (Niu, 2012)

sensors

Tea categories MOS Normalization, PCA; (Chen, et al., 2013; Dutta, et

SOM; FCM; al., 2003)

ANN(MLP; LVQ;
 ACCEPTED MANUSCRIPT

PNN; RBF)

Colorimetric sensors PCA; LDA (Chen, et al., 2013)

Black tea from MOS Normalization, PCA; (Tudu, Bhattacharyya, &

different gardens BP-MLP; Bikram, 2008; Tudu, Jana, et

in India incremental RBF al., 2009)

PT
neural network

Black tea MOS Normalization, PCA; (Bhattacharya, Tudu, Jana,

RI
fermentation SVD based 2NM and Ghosh, Bandhopadhyaya, &

process MDM; SOM; TDNN Bhuyan, 2008; Bhattacharyya,

SC
Seth, Tudu, Tamuly, Jana,

Ghosh, Bandyopadhyay, &

U
Bhuyan, 2007; Bhattacharyya,
AN
Seth, Tudu, Tamuly, Jana,

Ghosh, Bandyopadhyay,
M

Bhuyan, et al., 2007)

Quantitative Odor mixture’s MOSFET and MOS DWT; GA; SVR (Phaisangittisagul & Nagle,
D

analysis responses on 2011)

TE

machine olfaction

sensors
C EP
AC
 ACCEPTED MANUSCRIPT

Table 5 Summary of computer vision applications in tea analysis

Type of Model calibration

Property of study Technique of study References
study algorithm

Qualitative Tea categories Multi-spectral PCA; LSSVM; (Chen, Wu, & He, 2008; Li, He,

analysis imaging DCT; SVM; ANN; Qiu, et al., 2008; Li, He, Qiu, et

PT
LDA; BPANN al., 2007; Wu, et al., 2009)

RI
Tea grades Hyperspectral PCA; SVM; (Jiang, Qiao, Zheng, & Yang,

imaging C-SVM 2011; Zhao, Chen, Cai, &

SC
Ouyang, 2009)

Multi-spectral PCA; MSDA (Li & He, 2009)

U
imaging AN
Normal camera PCA; GNN (Laddi, et al., 2012; Wang, Zeng,

imaging & Du, 2010)

Tea varieties Normal camera PCA; LDA; SVM (Chen, Zhao, & Cai, 2008; Chen,
M

imaging based on RBF Zhao, Cai, & Wang, 2006)

D

Tea grains under Normal camera PCA (Laddi, Sharma, Kumar, & Kapur,

different imaging 2013)

TE

illumination

conditions
EP

Tea brands Multi-spectral WT; GLCM; (Li, He, & Qiu, 2008)

imaging STEPDISC
C

Tea color Normal camera SIMCA (Chen, Zhao, Zhang, & Fang,
AC

imaging 2005)

Quantitative Chlorophyll Hyperspectral MSAVI2 (Zhao, et al., 2011)

analysis content and imaging

distribution in tea

plant's leaves
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 Highlights
ACCEPTED MANUSCRIPT

Highlights

Conventional analytical methods to tea quality and nutrition are reviewed.

Applications of green analytical tools to tea quality and nutrition are reviewed.

Strengths of limitations of green analytical tools in tea are discussed.

PT

Related data processing algorithms for green analytical tools are discussed.

RI
U SC
AN
M
D
TE
C EP
AC