Food Research International 61 (2014) 2332

Contents lists available at ScienceDirect

Food Research International

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

Review

Application of infrared spectral techniques on quality and compositional

attributes of coffee: An overview
Douglas Fernandes Barbin a,, Ana Lucia de Souza Madureira Felicio a, Da-Wen Sun b,
Suzana Lucy Nixdorf c, Elisa Yoko Hirooka a
a
Department of Food Science, Center for Agricultural Sciences, State University of Londrina, Rodovia Celso Garcia Cid, PR 445 Km 380, Campus Universitrio, Londrina, PR 86055-900, Brazil
b
Food Refrigeration and Computerised Food Technology (FRCFT), School of Biosystems Engineering, University College Dublin, National University of Ireland, Agriculture & Food Science Centre,
Beleld, Dublin 4, Ireland
c
Department of Chemistry, Center for Exact Sciences, State University of Londrina, Rodovia Celso Garcia Cid, PR 445 Km 380, Campus Universitrio, Londrina, PR 86055-900, Brazil

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

Article history: During the last two decades, near and mid-infrared spectral analyses have emerged as a reliable and promising
Received 25 November 2013 analytical tool for objective assessment of coffee quality attributes. The literature presented in this review clearly
Received in revised form 23 December 2013 reveals that near and mid-infrared approaches have a huge potential for gaining rapid information about the
Accepted 3 January 2014
chemical composition and related properties of coffee. In addition to its ability for effectively quantifying and
Available online 11 January 2014
characterising quality attributes of some important features of coffee such as moisture, lipids and caffeine con-
Keywords:
tent, classication into quality grades and determination of sensory attributes, it is able to measure multiple
Coffee chemical constituents simultaneously avoiding extensive sample preparation. Developing a quality evaluation
Infrared spectroscopy system based on infrared spectral information to assess the coffee quality parameters and to ensure its authen-
FT-NIR tication would bring economical benets to the coffee industry by increasing consumer condence in the quality
Arabica of products. This paper provides an overview of the recently developed approaches and latest research carried
Robusta out in near and mid-infrared spectral technology for evaluating the quality and composition of coffee and the
Caffeine possibility of its widespread deployment.
2014 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2. Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.1. Spectra pre-processing and multivariate data analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1. Prediction of coffee composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2. Authentication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3. Classication of samples according to coffee variety and quality features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4. Discrimination between defective and non-defective samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5. Prediction of sensory properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6. Degree of roasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7. Coffee residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4. Future trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1. Introduction

Coffee is the world's most popular beverage after water, and one of
Corresponding author. Tel.: +55 43 33715963. the most important raw materials within the international trade for
E-mail address: [email protected] (D.F. Barbin). which quality is quintessential. Interest in coffee quality assessment is

0963-9969/$ see front matter 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodres.2014.01.005
24 D.F. Barbin et al. / Food Research International 61 (2014) 2332

impelled by the need to supply the consumer with a consistently high steel cup of approximately 10 mm in diameter, and the sample surface
quality product at an affordable price. Indeed, quality is a major aspect is attened. The cup is placed in a DRIFT accessory, which incorporates
for the modern coffee industry because a high quality product is the suitable mirrors to steer the infrared beam onto the sample, to collect
basis for success in today's particularly competitive market. the portion that is diffusely reected from the sample surface and to
Among the several species of the genus Coffea identied so far, two direct it onto the detector to record the spectrum. For certain samples,
of these varieties are economically and commercially important: Coffea dilution in another matrix, for example, powdered infrared grade potas-
arabica (arabica coffee) and Coffea canephora (robusta coffee) (Huck, sium bromide (KBr), is essential to avoid optical distortion effects in the
Guggenbichler, & Bonn, 2005; Illy & Viani, 1996). Both coffee varieties spectra (Kemsley et al., 1995). Essentially, spectroscopy provides com-
differ mutually from a botanical perspective and in terms of quality fea- prehensive information of biological samples as an effect of the interac-
tures. Arabica beans are more valued by the trade because they are con- tion between electromagnetic radiation and the sample material, such
sidered to have a ner avour and therefore more appreciated by the as reectance, transmittance, absorbance, phosphorescence, uores-
consumers than robusta. At present, most commercially available coffee cence, and radioactive decay (Abbott, 1999; Blanco & Villarroya, 2002).
beverages are produced from arabica and robusta roasted beans or Near-infrared spectroscopy (NIRS) is based on molecular overtones
blends of these two varieties (Esteban-Dez et al., 2004a). This raises and combination vibrations that can be very useful in probing bulk ma-
the possibility of fraudulent or accidental mislabelling, a subject of con- terial with little or no sample preparation (Huck et al., 2005; Pasquini,
cern to food processors and regulatory authorities (Downey & Boussion, 2003). The sensitivity of spectroscopic measurements relies on band
1996). Thus, it is important that the varieties of raw beans and of various intensities. Spectral information is repeated throughout the successive
coffee products can be properly identied. A trained inspector can easily overtones and combination regions; however, the signals of bands
distinguish raw arabica and robusta beans by differences in size and col- involved become weaker by an order of magnitude. The third over-
our. However, after roasting and milling, these visual indicators are tones encountered in shorter wavelengths such as the visible re-
eliminated, and identication of ground roast coffees requires alterna- gion (400800 nm) have considerably weaker band intensities
tive procedures (Kemsley, Ruault, & Wilson, 1995). when compared to the second and rst overtones in the NIR region
Chemical constituents of the roasted beans determine the quality of (8001800 nm) (Osborne, Fearn, & Hindle, 1993; Weyer, 1985). Al-
coffee as a beverage. Raw coffee beans contain a wide range of different though some unique molecule vibration can be properly assigned in
chemical compounds, which react and interact amongst themselves at the NIR region of the spectra, the complex composition of organic sam-
all stages of coffee roasting, resulting in greatly diverse nal products ples such as coffee, where over 800 compounds ranging from simple
(Ribeiro, Ferreira, & Salva, 2011). For instance, caffeine content which linear and branched structures to complex cyclic and heterocyclic struc-
has a signicant effect on the nal quality of the coffee products needs tures can be encountered, makes it difcult to analyse (Illy & Viani,
to be determined fast and reliably by analytical techniques. Due to the 1996; Lyman, Benck, Dell, Merle, & Murray-Wijelath, 2003; Wang, Jun,
high number of samples to be analysed, the coffee industry needs new Bittenbender, Gautz, & Li, 2009).
analytical techniques providing fast and reliable data about coffee The literature on chemical assignments for some of the most impor-
quality. tant NIR bands is summarized in a schematic representation of spectra
Over the past few years, a number of methods have been investigat- structure correlation in Fig. 1. In addition, the wavelength range in
ed to measure coffee quality traits in objective ways (Bicchi, Ombretta, which have been reported the greatest absorbance of pure components
Pellegrino, & Vanni, 1997; Farah, Monteiro, Calado, Franca, & Trugo, of coffee are presented.
2006; Franca, Mendona, & Oliveira, 2005; Franca, Oliveira, Mendona, Applications of infrared spectral analyses have increased in food
& Silva, 2005; Gonzalez, Pablos, Martin, Leon-Camacho, & Valdenebro, product quality analysis, and it has been widely used to predict the attri-
2001; Mendona, Franca, & Oliveira, 2009; Mendona, Franca, Oliveira, butes of grains (Jing et al., 2010), tea (Chen, Zhao, Zhang, & Wang, 2006;
& Nunes, 2008; Perrone, Donangelo, & Farah, 2008). Nowadays, the cof- Schulz, 2004), honey (Chen et al., 2011), fruits and vegetables (Bauer
fee industry has a growing need for consistent reference methods for et al., 2008; Pedro & Ferreira, 2005; Xie, Ying, Ying, Yu, & Fu, 2007;
developing robust and accurate calibration models, which can further Yan-De et al., 2007), beverages (Cozzolino, Smyth, & Gishen, 2003; He,
be used in the determination of additional samples (Esteban-Dez Rodriguez-Saona, & Giusti, 2007; Inn, Garrigues, & Guardia, 2006),
et al., 2004a). Spectroscopy is an increasingly growing technique due oils (Downey, Mcintyre, & Davies, 2003), dairy products (Downey et al.,
to its rapidity, simplicity, and safety, as well as its ability to measure 2005; Fagan et al., 2007; Woodcock, Fagan, O'Donnell, & Downey,
multiple attributes simultaneously without monotonous sample prepa- 2008), eggs (Abdel-Nour, Ngadi, Prasher, & Karimi, 2011), meat (De
ration. More specically, spectroscopic methods in the visible, near and Marchi, 2013; Liao, Fan, & Cheng, 2012), and other agricultural products,
mid-infrared ranges are rapid, chemical-free and low cost techniques in order to substitute other commonly used destructive methods. It is
without the need of sample processing that has been widely used to de- widely used to quantify the composition of agricultural products because
tect the chemical composition of food materials, making it suitable to be it meets the criteria of being accurate, reliable, rapid, non-destructive,
implemented on a routine basis (Jing, Cai, & Shao, 2010; Pedro & and relatively inexpensive. Fourier transform infrared (FTIR) has also
Ferreira, 2005; Santos, Sarragua, Rangel, & Lopes, 2012; Yan-De, Yi- been investigated regarding coffee applications, although this technique
Bin, Xiaping, & Guisan, 2007). is more sensitive to the heterogeneity of the individual coffee samples,
This paper thus provides an overview of the application of infrared because stretching and deformation vibrations are directly used and
spectral technique as a reliable procedure compared to traditional ana- not their overtones or combination vibrations.
lytical methods to assess coffee composition and quality parameters and
classify coffee samples from different varieties and quality grades. 2.1. Spectra pre-processing and multivariate data analyses

2. Spectroscopy Certain effects such as light scattering that occurs in diffuse reec-
tance NIR spectroscopy are inuenced by sample presentation causing
Near and mid-infrared spectral techniques have received consider- unexpected perturbations in spectra (baseline shifts, slope changes
able attention among researchers as a tool for rapid, non-destructive and non linearity), generating spectral changes that are not related to
and simultaneous assessment of multiple attributes of organic materials the studied responses and affecting the reliability of multivariate cali-
without lengthy sample preparation. While near infrared (NIR) spec- bration models (Osborne et al., 1993). As a result, NIR signals contain
troscopy requires almost no sample preparations, obtaining a diffuse re- information which is not only from chemical but also physical charac-
ectance infrared Fourier transform (DRIFT) spectrum requires some teristics from samples and measuring conditions. Since wavelength de-
prudence. A powdered or ground sample is placed into a stainless pendency of light scatter is different from that of chemically based light
 D.F. Barbin et al. / Food Research International 61 (2014) 2332 25

Fig. 1. Schematic representation of spectrastructure correlation and absorption regions of main components found in coffee (adapted from Osborne et al., 1993; Ribeiro et al., 2011).

absorption, scattering effects can be attenuated by some mathematical articial neural networks (ANN) are often employed to extract the
treatments, which include derivation, multiplicative scatter correction desired chemical information. Other multivariate techniques that have
(MSC), standard normal variate (SNV) and orthogonal signal correction found applicability in spectral analyses include genetic algorithm
(OSC) (Esteban-Dez et al., 2004a). Therefore, the application of suitable (GA), soft independent modelling of class analogies (SIMCA), and suc-
pre-processing methods, aimed at correcting spectral data by minimiz- cessive projection algorithms (SPA), among others (Zhang et al.,
ing the contribution of physical effects to NIR signals is a crucial step in 2013). Careful development of a set of calibration samples and applica-
regression model development. tion of multivariate calibration techniques are essential for near-
Additionally, one of the characteristics inherent to spectroscopic infrared analytical methods (Isaksson & Naes, 1988).
analysis is the number of variables (several hundreds or even thou-
sands) that exist in the NIR spectra, usually signicantly larger than 3. Applications
the number of samples. For this reason, they contain interferences com-
ing from background, noise and overlapping bands, leading to redun- The feasibility of NIR techniques as an alternative to reference
dancy and co-linearity in these variables. The molecular overtone and methods for coffee classication and authentication, and prediction of
combination bands seen in the NIR range are typically very broad, lead- coffee composition have been investigated. Most recent applications of
ing to complex spectra; thus being a difcult task to assign particular NIR spectroscopy to coffee samples are listed in Table 1. Several spectral
features to specic chemical components. It brings a great challenge to pre-processing techniques and multivariate analyses were tested. Rele-
build a high-quality prediction model for unknown set of samples. vant observations and results are further presented and discussed.
Thus, the application of an efcient variable selection and efcient sig-
nal de-noising and compression methods prior to the calibration step 3.1. Prediction of coffee composition
has been proved to be greatly benecial in providing more reliable
and parsimonious models (Pizarro, Esteban-Diez, & Gonzlez-Siz, Several works have investigated the potential application of spectral
2007; Telo, Martins, & Ferreira, 2009; Zhang et al., 2013). techniques to assess physical, chemical and quality parameters of coffee.
Multivariate (multiple variables) calibration techniques such as Most of the investigations have focused on developing an alternative to
principal components analysis (PCA), partial least squares (PLS), or reference methods for coffee attribute prediction and to classify coffee
26 D.F. Barbin et al. / Food Research International 61 (2014) 2332

Table 1
Most recent applications of IR spectral analysis in coffee.

Application Task Wavelength range Statistical analysis Author(s)/year

Adulteration/ Presence of glucose, starch, chicory 8004000 cm1 (250012,500 nm) FT-IR LDA, PCA, PLS, ANN Briandet, Kemsley, & Wilson (1996)
authentication Presence of barley 10,0004000 cm1 (10002500 nm) FT-NIR PLS and PLSGA Ebrahimi-Najafabadi et al. (2012)
Robusta content in roasted coffee 11002500 nm PLS Pizarro, Esteban-Diez, & Gonzlez-Siz (2007)
Authentication of Kona coffee 4000400 cm1 (250025,000 nm) FT-IR PLS and PCR Wang et al. (2009)
Classication/ Samples from different batches 8002500 nm PLS-DA and SIMCA Santos, Moura, et al. (2012)
discrimination and producers
Identifying arabica and robusta 11002500 nm Potential functions Esteban-Dez, Gonzlez-Siz and Pizarro (2004a)
methods
Classication into arabica and 11002500 nm Potential functions Esteban-Dez, Gonzlez-Siz, Saenz-Gonzalez,
robusta methods & Pizarro (2007)
Identifying arabica and robusta 4002498 nm (PCA) and FDA Downey & Boussion (1996)
Identifying arabica and robusta 4002498 nm/4000800 cm1 FDA and PLS Downey, Briandet, Wilson, & Kemsley (1997)
(250012,500 nm)
Arabica and robusta 8004000 cm1 (250012,500 nm) FT-IR PCA and DA Kemsley et al. (1995)
Discrimination/ Defective and non-defective beans 4000700 cm1 (250014,285 nm) FT-IR PCA, clusters Craig, Franca, & Oliveira (2012a)
prediction Defective and non-defective beans 4000400 cm1 (250025,000 nm) FT-IR PCA and LDA Craig, Franca, & Oliveira (2012b)
Defective beans 10,0004000 cm1 (10002500 nm) FT-IR PCA and PLS Santos, Sarragua, Rangel, & Lopes (2012)
Composition Caffeine content 8004000 cm1 (250012,500 nm) FT-IR Linear regression Garrigues, Bouhsain, Garrigues, & de la Guardia
(2000)
Caffeine, theobromine and 99964008 cm1 (10002495 nm) FT-NIR PLS Huck et al. (2005)
theophylline
Moisture content 450010,000 cm1 (10002222 nm) PLS Morgano, Faria, Ferro, Bragagnolo, & Ferreira
(2008)
Ash and lipid content 11002500 nm PLS Pizarro, Esteban-Diez, Nistal, & Gonzlez-Siz
(2004)
Caffeine content and roasting 11002500 nm PLS Pizarro, Esteban-Diez, Gonzalez-Saiz, & Forina
colour (2007)
1
Caffeine content 400010,000 cm (10002500 nm) PLS Zhang et al. (2013)
Sensory prediction Sensory attributes 4000400 cm1 (250025,000 nm) FT-IR PCA and PLS-DA Ribeiro, Salva, & Ferreira (2010)
Sensory attributes 11002500 nm PLS Ribeiro et al. (2011)
Sensory properties 11002200 nm PLS (and IPWPLS) Esteban-Dez, Gonzlez-Siz, & Pizarro (2004c)
Degree of roasting Prediction of roasting degree 400012,000 cm1 (8302500 nm) FT-NIR PLS Alessandrini, Romani, Pinnavaia, & Rosa (2008)
Effect on roasting conditions 4000700 cm1 (250014,285 nm) FT-IR Lyman et al. (2003)

according to physico-chemical characteristics into coffee varieties or most important bands were mainly related to the vibration of O\H and
origins. C\H bonds between 4650 and 4900, 5004 and 5062 cm1. The regres-
Water content above 12.5% in green coffee causes a number of sion model allowed for the determination of moisture content of raw cof-
unwanted consequences like microbial growth, mycotoxin formation, fee samples with a standard error of validation of 0.298 g 100 g1 and
altered sensorial quality of nal product, and unstable production con- correlation coefcient (r) of 0.818 for validation, with an average relative
ditions among others. Several standards for reference, routine and error of 4.1% (Morgano et al., 2008).
rapid methods are already established for determination of water con- Caffeine content is an important component of coffee and thus has
tent in green coffee. Nevertheless, there are discussions on how effec- been the target of several investigations. Ground arabica coffee samples
tive the methods are in determining the water content. The work of within a wide range of roasted levels were analysed by NIR in order to
Reh, Gerber, Prodolliet, and Vuataz (2006) focused on clarifying the obtain a reduced-spectrum model for quantitative determination of caf-
specicity and accuracy of several available reference methods for feine content. PLS regression was used to construct calibration models
determination of water content in green coffee, demonstrating that based on chemometric analyses of the NIR spectral data and caffeine
only ISO 1446 exclusively measures water. For all drying oven based contents determined by the high performance liquid chromatography
methods it was observed that degradation of the product contributes with ultraviolet (HPLCUV) detection method as the reference values.
to the overall weight loss. Near-infrared spectroscopy and colour mea- Comparing the respective quality of the different models studied, the
surement were applied to study the different drying methods, establish- application of a second derivative pre-treatment and stability competi-
ing the degree of degradation and the completeness of the drying tive adaptive reweighted sampling (SCARS) variable selection provided
process. The drying process was straightforwardly monitored due to a notably improved regression model, with a root mean square error of
the specicity of the 1940 nm wavelength band for water molecules prediction (RMSEP) of 0.378 mg/g, mean relative error of 1.98% and
by comparing the weight loss results with the residual NIRS absorption mean relative standard deviation (RSD) of 1.71%. Thus, the results re-
left after drying. It was observed that neither reference methods ISO vealed the feasibility of NIR spectroscopy for at-line application to pre-
1446 nor ISO 6673 allows complete drying of green coffee, since ISO dict the caffeine content of unknown roasted coffee samples, thanks to
1446 has proved to let residual water content at the end of the drying the short analysis time of a few seconds and non-destructive advantages
process (e.g., 2 weeks). This residual value was very sensitive to the par- of NIRS (Zhang et al., 2013).
ticle size, due to the very low water diffusion in dehydrated green coffee Regarding caffeine content, a fast procedure was proposed for the
material. All methods discussed (ISO 1447; ISO 6673; ISO 1446) were FTIR determination of this component in roasted coffee samples with re-
shown to be highly correlated with each other with R2 values above duced amount of organic solvent used. The method involves extracting
0.99. the caffeine with CHCl3, and measuring absorbance at 1659 cm1 using
On another study, NIR spectroscopy was investigated to measure the a baseline established between 1900 and 830 cm1. The procedure pro-
moisture content in raw coffee. The original NIR spectra were pre- vides a drastic reduction of the organic solvent consumed by the refer-
processed through different transformations and mathematical pre- ence chromatographic UVspectrometric determination, with a limit of
treatments, such as the KubelkaMunk transformation; multiplicative detection of the order of 3 mg L1 of caffeine. The accuracy of the FTIR
signal correction (MSC); spline smoothing and movable average. The procedure was higher than 94% (Garrigues et al., 2000).
 D.F. Barbin et al. / Food Research International 61 (2014) 2332 27

Quantitative calibration models were developed using diffuse reec- coffee present on the market. Furthermore, four types of barley samples,
tance NIR spectra of liquid coffee beverages as predictors collected by a with different roasting degrees, were blended to coffee in concentra-
FT-NIR spectrometer. Determination coefcients (R2) for the calibration tions between 2 and 20% weight per weight of barley, with the low-
models of instant coffee, plant fat and sugar, were 98.97%, 99.94% and est limit being surely lower than the sensorial limit of detection, and
99.18% respectively; with root mean square errors of cross validation resolution step (2%) lower than human sensorial capability for
(RMSECV) of 2.12, 0.72 and 2.01 respectively, indicating that NIR spec- distinguishing between close quantities. A genetic algorithm (GA)
troscopy can be applied for the simultaneous determination of the three was applied in order to obtain simplied models, taking into account
main ingredients in liquid coffee beverage (Wang, Min, Duan, Xiong, & only informative regions of the spectral proles. Very low prediction
Li, 2012). errors were achieved on an external test set, revealing that NIRS can
An analytical method based on NIRS for the quantication of alka- nd applications to quantify adulterants in coffee powder. However,
loids (caffeine, theobromine and theophylline) was investigated in it must be emphasized that sample collection and analysis should be
roasted coffee after discrimination of the rough green beans into arabica performed through a number of years, in order to account for a wider
and robusta. This validated method was compared to the most com- range of variability factors and to obtain models for a large-scale ap-
monly used liquid chromatography (LC) connected to UV and mass plicability (Ebrahimi-Najafabadi et al., 2012). The promising results
spectrometric (MS) detection. Among the most important spectral fea- obtained can be considered in future applications to quantify differ-
tures, it was noticed that the vibration of the 2nd overtone of the ent blends of varieties in order to identify fraudulent mixtures.
carbonyl group (5352 cm1), C\H stretch and C\H deformation vibra- Among other types of coffee adulteration, the most commonly en-
tion, and the CH2 overtone (5742 cm 1), was caused by ingredients countered are adulteration with coffee substitutes such as chicory,
such as lipids, volatile and non-volatile acids, chlorogenic acids, alka- malt, gs, cereals, caramel, starch, maltodextrins or glucose, as well as
loids and some aroma compounds. Theobromine has one methyl roasted or even unroasted coffee husks; mixing of two species (addition
group less than caffeine and showed differences compared to caffeine of cheaper robusta to pure arabica coffee); or mixing of expensive coffee
at 7353 cm1 (C\H str. +C\H def.), 5865 cm1 (C\H str., rst over- beans from one growing region with cheap beans grown in another re-
tone) and 4383 cm1 (C\H str. + C\H def.). Analysis of 83 liquid cof- gion. Most used methods for detection of coffee adulteration are gener-
fee extracts resulted in correlation coefcients of 0.86 and 0.85 in a ally based on chemical parameters (Franca, Oliveira, Mendona, & Silva,
concentration range between 0.10 and 4.13 g.100 g1, with values for 2005; Gonzalez et al., 2001; Mendona et al., 2009). Infrared spectrosco-
standard error of prediction (SEP) of 0.07 and 0.10 g 100 g1 for caf- py has been explored as an alternative to existing methods for the de-
feine and theobromine, respectively. The robustness and reproducibility tection of undeclared material in instant coffee. Two different FTIR
of the NIRS model for the determination of caffeine and theobromine sampling methods were tested: diffuse reectance, and attenuated
showed that it can be used to predict the content of these components total reectance. PCA and linear discriminant analysis (LDA) were ap-
in liquid extracts, with the possibility for the coffee producing industry plied to distinguish between pure samples and samples adulterated
to replace the LC method usually applied to determine these compo- with glucose, starch or chicory in the range 20100 g kg1, achieving
nents in the routine analysis (Huck, Maurer, Popp, Basener, & Bonn, a 98% successful classication rate. Simultaneous PLS regression was
1999). However, due to the lower limit of detection of the NIRS- carried out for the content determination of three added carbohydrates
method, the analysis of theophylline by LC method has to be favoured (xylose, glucose and fructose). An articial neural network was used to
due to the sensitivity of the method (Huck et al., 2005). classify pure and adulterated instant coffee based on spectral informa-
tion, attaining 100% of correctly classied samples. The results present-
3.2. Authentication ed show that the different statistical approaches yield complementary
information. The PCA gives a visual representation of the position of
Food authentication is a major issue that has become increasingly samples in a low-dimensional space, and the LDA uses the relative posi-
important in recent years. The need to guarantee the actual origin of a tions of samples in this space to form a useful classication rule. The
product and to determine whether it has been mixed with adulterants ANN analysis offered an improvement over the classication results ob-
is a major challenge to the food industry, regulatory authorities and con- tained by LDA, while the PLS regressions for the added xylose, glucose
sumers, which are all interested in authenticating raw materials and and fructose contents show that FTIR spectroscopy has the required
food products in order to satisfy food quality and safety requirements sensitivity for determining the carbohydrate prole of samples
(Pizarro, Esteban-Diez, & Gonzlez-Siz, 2007). (Briandet et al., 1996).
The excellent quality of Kona coffee, the variety of Kona typica
grown in Kona-Island (Hawaii, U.S.A.), carries a unique stamp and is
among the most distinctive coffee products in the world. Fourier trans- 3.3. Classication of samples according to coffee variety and quality
form infrared (FTIR) spectroscopy integrated with an attenuated total features
reectance (ATR) accessory and multivariate analysis were used to de-
velop an effective analytical tool to discriminate pure Kona coffee and its Generally, arabica coffee beans are considered superior to robusta
adulterants and to quantify the percentage of Kona coffee in Kona coffee and are therefore the more expensive of the two. As a result, the identi-
blends. A calibration model was successfully validated using nine blend cation of adulterations and mislabelling is very important regarding
sets of 100% Kona coffee mixture and its adulterant, 100% non-Kona coffee the consumer protection (Huck et al., 2005). The potential of NIR reec-
mixture. Distinct peak variations of ground and brewed coffee blends in tance spectroscopy for discriminating between arabica and robusta cof-
the spectral region were observed between 800 and 1900 cm1. A PLS- fees and characterization of blends between these two varieties has
2nd derivative calibration model based on brewed Kona coffee with been investigated with relative success. Downey et al. (1997) compared
mean centering data processing showed the best predictability with the the near- and mid-infrared spectra of a set of 56 lyophilized coffee sam-
lowest standard error of calibration value of 0.81 and the highest R2 ples to discriminate between arabica and robusta lyophilized coffees.
value of 0.99 (Wang et al., 2009). Separate mathematical models were used for distinct wavelength
Regarding the identication of different coffee blends and authenti- ranges used as predictors separately and combined. Varietal authentica-
cation issues, an investigation based on NIR spectral information has tion of the coffees was tested using factorial discriminant analysis (FDA)
been carried out for the identication and quantication of the fraudu- and PLS. The combination of the two wavelength ranges required fewer
lent addition of barley in roasted and ground coffee samples. Nine types loadings or principal components for both FDA and PLS approaches test-
of commercial coffee samples, including pure arabica, robusta and mix- ed than with either spectral range alone, suggesting that a broader spec-
tures of them were chosen to extensively explore the variability of tral range may have advantages for this particular type of assessment.
28 D.F. Barbin et al. / Food Research International 61 (2014) 2332

On another study, dried beverages were produced by both measurements, proving its reliability and effectiveness to assess the
lyophilisation and air-drying under vacuum on a glass-bre lter paper, genuineness of coffee samples.
and spectral information in the wavelength range of 11002498 nm Orthogonal wavelet correction (OWAVEC) pre-processing method
was used for classication into coffee varieties. Identication of spectral combined with PLS regression was applied in order to simultaneously
features from bonding structures observed at 1672, 2158, 2250, 2332 operate two crucial pre-processing steps in multivariate calibration: sig-
and 2416 nm was related with features in caffeine, implying the differ- nal correction and data compression. Several pre-processing methods
ence in caffeine between dried arabica and robusta coffee beverages. In (mean centering, rst derivative, OSC and DOSC) were additionally ap-
addition, peaks related to water at 1454 and 1930 nm were observed plied in order to nd calibration models with the best predictive ability
inuencing the analyses. Correct classication rates of 87% and 95% and to evaluate the performance of the OWAVEC method, comparing
were achieved on lyophilised and vacuum-dried beverage samples re- the respective quality of the different regression models. However, the
spectively, supporting its application for authentication tasks. It was ob- success of this methodology was highly dependent on signal pre-
served that the basis for this discrimination appears to involve caffeine processing methods applied to minimize the spectral variation not
and/or other alkaloids, thus indicating that the quantication of these due to the parameter of interest but due to variation in experimental
components is a relevant aspect for coffee analyses (Downey & or sample conditions. The regression models developed for predicting
Boussion, 1996). robusta content of blends, after pre-processing rst derivative NIR spec-
In an attempt to improve the classication models constructed using tra by OWAVEC, were accurate (0.79% RMSEP) and considerably better
original roasted coffee NIR spectra, orthogonal signal correction (OSC) compared to the models obtained from raw data (in terms of both reli-
and direct orthogonal signal correction (DOSC) methods were applied ability and parsimony) and improved results were also obtained with
on the raw spectra. The proposed method aimed for removing informa- respect to the other orthogonal signal correction methods tested. The
tion not related to an explicitly chosen chemical variable that could pro- high quality calibration models constructed suggest that the proposed
vide discrimination between arabica and robusta coffees. Caffeine methodology would be a suitable tool for the quantitative detection of
content was preferred as this specic descriptor from among other robusta addition to arabica coffees. However, the authors state that al-
chemical parameters (chlorogenic acid and total acidity), since it though it was a feasibility study and further research should be carried
showed a notable discrepancy between varieties. The results obtained out before its value for authentication purposes. The data set used in this
before and after transforming the spectra were analysed and compared work was designed to cover, as much as possible, the great natural var-
in order to evaluate the effect of orthogonal corrections. It was demon- iability inherent in commercially available coffees by considering differ-
strated that the application of an orthogonal signal correction method ent roasting conditions and degrees, and varied geographical origins.
induced a substantial improvement of the classication models in com- However, it is important to emphasize that the dynamic nature of the
parison with the model developed on the basis of original spectra, even coffee market would require a more exhaustive collection of calibration
though classication models developed from unprocessed NIR spectra samples to serve as the basis for developing a predictive model for use in
provided satisfactory classication results. Moreover, the results obtain- the routine quality control of coffee samples (Pizarro, Esteban-Diez, &
ed using the methodologies proposed were as good as those obtained Gonzlez-Siz, 2007).
with the classication model constructed using a chemical descriptor Analogous to the situation found in calibration, a classication
in terms of both classication and prediction abilities (Esteban-Dez model constructed from spectra measured on one instrument may not
et al., 2004a). On another study, FTIR spectroscopy in the mid-infrared be valid for prediction of class from spectra measured on a second in-
region (8004000 cm1) combined with the diffuse reectance infra- strument. Thus, the transfer of multivariate classication models be-
red Fourier transform spectroscopy (DRIFTS) sampling technique and tween different near-infrared spectrometers was investigated for the
PCA revealed clear grouping of the spectra for discrimination of arabica discrimination of whole, green arabica and robusta coffee beans. Orthog-
and robusta ground roast coffees. Classical discriminant analysis onal signal correction trained on a vector of discrete class identities was
(DA) based on principal components scores yielded 100% successful found to perform well in the pre-processing of data allowing the trans-
discrimination. However, this work did not report the identication fer of a classication model developed on data acquired on one instru-
of blends of different concentrations from the two varieties (Kemsley ment to be used on another instrument, resulting in robust models for
et al., 1995). the discrimination of green coffee beans on both spectrometers with
Nevertheless, although results have shown the feasibility of using misclassication errors in the range of 510% (Myles, Zimmerman, &
original NIR spectra of coffee samples to develop classication models Brown, 2006).
with good discrimination abilities between pure coffee varieties, it Other studies have reported the ability of NIR spectroscopy in
would be interesting to avoid potential errors when working with dif- discriminating coffees from different qualities. Santos, Moura, et al.
ferent coffee blends. Quantifying the robusta variety content of roasted (2012) described the use of near infrared spectroscopy (NIRS) and che-
coffee samples, as a means for controlling and avoiding coffee adultera- mometric techniques for the classication of coffee samples from differ-
tion, is a very important issue taking into account the great variability of ent lots and producers acquired in supermarkets and roasting industries
the nal sale price depending on coffee varietal origin (Pizarro, Esteban- in Brazil. In general, it was found that NIR spectroscopy, coupled with ei-
Diez, & Gonzlez-Siz, 2007). ther SIMCA or partial least squares discriminant analysis (PLS-DA) mul-
Esteban-Dez et al., 2007 proposed an innovative approach capable tivariate models, can be a useful tool to differentiate roasted coffee
of discriminating between arabica and robusta pure coffee samples grains.
and blends of these two varieties. The results reported have demon-
strated that the combination of near-infrared spectroscopy with the 3.4. Discrimination between defective and non-defective samples
OSC method and with a class-modelling technique, such as potential
functions method, yielded better results than using the raw spectra. Characterisation of coffee quality based on bean quality assessment
Therefore, this approach can be used as an improved strategy not only is associated with the relative amount of defective beans among non-
for discriminating between arabica and robusta pure coffee varieties, defective ones. It is therefore important to develop a methodology
but also for differentiating between blends of the two species, even that enables a fast assessment of coffee grade and capable of becoming
when separate categories of blends are dened depending on the actual an analytical tool to regulate coffee quality. A methodology for quality
robusta content in samples. The applied strategy did not imply any addi- assessment of green coffee based on near infrared spectroscopy
tional reference analysis, since the response variable was precisely the (NIRS) was proposed for fast assessment of arabica and robusta coffee
percentage of robusta coffee in each sample already known for calibra- varieties from different geographical locations and identication of the
tion. Therefore, the applied classication strategy only relied on NIR presence of defective beans in a batch. PLS regression relating the NIR
 D.F. Barbin et al. / Food Research International 61 (2014) 2332 29

spectrum to the mass fraction of defective and non-defective beans pre- hypothesis that a low number of NIR absorption bands corresponding
sented relative errors around 5% for four coffee provenances investigat- to certain compounds strongly related to a specic attribute can predict
ed (two robusta, Indonesia and Vietnam, and two arabica, Colombia and it with similar or even improved accuracy and reliability as using the
Nicaragua). The results obtained are an indication that although the de- whole NIR wavelength range. Specic scores obtained for each espresso
fects could not be recognized individually it could be possible to quanti- sample in the sensory analysis were the direct consequence of the par-
fy the non-defective quality against the remaining qualities, showing ticular chemical composition of the roasted coffee used as raw material.
that NIRS enables a fast, simple and quantitative evaluation of green cof- Thus, separate PLS regression models were constructed based on the
fee following general international guidelines of coffee classication corresponding NIR roasted coffee spectra in order to model each senso-
(Santos, Sarragua, Rangel, & Lopes, 2012). ry property studied. The results provided by the calibration models
The feasibility of employing DRIFTS for discrimination between were comparable in terms of accuracy to the evaluations provided by
defective and non-defective coffees after roasting and grinding has a trained sensory panel, proving the feasibility of using such methodol-
shown promising results. PCA applied to the spectra and of the rst de- ogy in on-line or routine applications (Esteban-Dez et al., 2004c).
rivatives of the spectra provided separation of the samples into four
groups: non-defective, black, dark sour and light sour, with immature 3.6. Degree of roasting
beans scattered among the sour samples. Classication models based
on LDA ranged from 95 to 100% (Craig et al., 2012b). On another Roasting colour and quality parameters have been targeted as main
study, Craig et al. (2012a) evaluated the potential of FTIR spectroscopy attributes investigated by NIR spectroscopy (Esteban-Dez, Gonzlez-
combined with PCA and cluster analysis to discriminate defective and Siz, & Pizarro, 2004b). For instance, FTIRATR was used to study the ef-
non-defective beans. Results have shown the separation of the samples fect of dark, medium and light roasting conditions on coffee brewed
into two major groups: non-defective/light sour and black/dark sour/ from Guatemala Antigua arabica coffee beans. The 18001680 cm 1
immature, indicating that non-defective and defective (black, immature carbonyl region of the spectra provided a avour-print which appears
and sour) coffee samples could be separated into distinct groups based to be consistent with the taste and aroma perceived by sensory
on transmittance or reectance spectra. It is indicated that FTIR presents panellists. The composition of some carbonyl compounds (vinyl esters/
potential for the development of a fast and reliable analytical methodol- lactones, esters, aldehydes, ketones, and acids) in the brewed coffee ap-
ogy for discrimination between defective and non-defective coffee, al- pears to be dependent upon the heating rate of the green beans to the
though the authors suggest that further studies should be conducted onset of the rst and second cracks. It was observed that when the rate
employing larger sets of samples in order to develop more robust pre- of heating to the onset of the rst and second cracks was kept constant,
dictive models. It is emphasized, however, that all classes of compounds the types of carbonyl compounds formed were similar, varying only in
are evenly distributed throughout the bean, including its surface. There- their concentration. This difference in concentration is apparently due to
fore, when the beans are ground, the chemical makeup of the bean sur- the additional heating of the coffee bean beyond the second crack.
face will remain embedded in the sample and thus will contribute to the When the heating rate to the onset of the rst and second cracks was var-
makeup of the FTIR spectra. ied, both the types and concentration of the carbonyl compounds formed
during roasting were affected. Thus, heating rates of green coffee beans to
3.5. Prediction of sensory properties the onset of the rst and second cracks are important determinants of the
basic taste and aroma of brewed coffee (Lyman et al., 2003).
Further studies were carried out to establish a relationship between The potential of mid-infrared diffuse reectance spectroscopy with
the sensory attributes of the beverage and the chemical components of Fourier transform was investigated for discrimination of commercial
the coffee beans. Chemometric analyses based on coffee beverage sen- coffee samples. A clear discrimination of decaffeinated from medium
sory data and NIR spectra of arabica roasted coffee samples aimed to and dark roasted coffees was observed by application of PCA. A PLS-
predict the scores of acidity, bitterness, avour, cleanliness, body and DA model correctly classied 100% of the external validation and pre-
overall quality of coffee beverage determined by sensory panellists. diction samples according to their roasting degree. Chemometric analy-
PLS were used to establish prediction models and the ordered predictor ses of the MIR spectra allowed inferring on the lower carbohydrate,
selection (OPS) algorithm was applied as a primary step to select the caffeine and chlorogenic acid concentration as well as on the higher
wavelengths mostly related to each sensory attribute in order to select moisture in the decaffeinated coffee compared with traditional and
only signicant regions for the regression models. All models presented dark roasted coffees (Ribeiro et al., 2010).
good correlation (rcv) values for the respective attribute predicted thus The most common mathematical treatments applied to correct sys-
being useful for associating specic spectral regions to the spectra of the tematic variations unrelated to the studied responses are derivation
pure compounds. It was observed that the regions of the spectrum de- SNV and MSC. Other mathematical treatments, such as OSC and DOSC
ned as important for sensory quality were closely related to the NIR have been developed in order to minimize the variability unrelated to
spectra of pure caffeine, trigonelline, 5-caffeoylquinic acid, cellulose, the response in spectral data. It has been conrmed that the pre-
coffee lipids, sucrose and casein. The NIR analyses conrmed the rela- processing methods most usually applied (derivation, MSC, SNV) can
tionship between the sensory characteristics of the beverage and the reduce the variation unrelated to the modelled response, and that, in
chemical composition of the roasted grain, demonstrating that the lipids some cases, they lead to better calibration models compared to those
and proteins in the roasted bean were closely related to the attribute of obtained from the original spectra. However, in certain applications, it
body in the coffee beverage, caffeine and chlorogenic acids to bitterness was shown that none of these pre-treatments can remove completely
and chlorogenic acid to acidity, and the avour, cleanliness and overall all the systematic variability. By contrast, OSC and DOSC have proven
quality were related to the caffeine, trigonelline, chlorogenic acid, poly- their relative effectiveness to correct coffee spectra, both for the quanti-
saccharides, sucrose and protein present in the roasted coffee beans cation of ash content and for the determination of total lipids. These
(Ribeiro et al., 2011). transformations removed at least a portion of information unrelated
Application of NIR spectroscopy for predicting sensory attributes to the response, leading to signicantly improved and simpler calibra-
from an espresso quality assurance (perceived acidity, mouthfeel tion models that only require a few components to model the data
(body), bitterness and aftertaste) was studied. Wavelength selection and providing predictive abilities much higher than other methods.
was performed applying iterative predictor weightingPLS (IPWPLS) Compared to models obtained with the original data and the data
(Forina, Casolino, & Pizarro, 1999) in order to take into account only sig- corrected by derivation, SNV and MSC, OSC and DOSC treatments gave
nicant and characteristic spectral features to improve the quality of the PLS calibration models with improved prediction abilities (4.9 and
nal regression models constructed. The objective was to test the 3.3% RMSEP with corrected data versus 7.1 and 8.3% RMSEP with
30 D.F. Barbin et al. / Food Research International 61 (2014) 2332

original data, respectively). Therefore, both OSC and DOSC could be used process and for controlling coffee quality and possible quality changes
to develop reliable regression models based on ltered data for the spe- (Pizarro, Esteban-Diez, Gonzalez-Saiz, & Forina, 2007).
cic applications of this work (Pizarro et al., 2004).
The relationship between some coffee roasting variables (weight 3.7. Coffee residues
loss, density and moisture) with near infrared (NIR) spectra of original
raw (green) and differently roasted coffee samples was investigated in Spent coffee grounds are a signicant by-product which contains
order to predict coffee roasting degree. Using PLS regression, a predic- high levels of bioactive compounds, such as chlorogenic acid and avo-
tion of the three modelled roasting responses was performed. Most noids that have recognized antioxidant properties. Fourier-transform
signicant spectral differences were observed around 66007000, near infrared (FT-NIR) spectroscopy was used to determine the antiox-
54005900, 53005000, 48004500 and 44004100 cm1. Such spec- idant capacity, total phenolic and total avonoid contents of several
tral variations were likely related to changes in chemical composition spent coffee grounds samples. FT-NIR spectra were related to total a-
occurring during roasting, like the formation or degradation of certain vonoid and total phenolic contents and the antioxidant capacity using
compounds responsible for particular absorptions that strongly affect PLS modelling. Results were promising, with an R2 for an independent
the nal spectral prole. For instance, water band intensity at around test set of 0.93, 0.96, 0.95 and 0.95 for antioxidant capacity of spent cof-
6896 (rst overtone of O\H stretching) and 5154 cm1 (combination fee grounds, antioxidant capacity of spent coffee grounds ethanolic ex-
band of O\H stretching and O\H deformation) decreased gradually tracts, total avonoid and total phenolic contents, respectively. It was
during roasting. Obtained data allowed constructing robust and reliable conrmed that NIR spectroscopy has possible applications for routine
models for the prediction of roasting variables of unknown roasted cof- assessment of these parameters and is a viable and advantageous alter-
fee samples, since measured and predicted values showed high correla- native to chemical procedures involving laborious extractions (Pscoa
tion coefcients (r from 0.92 to 0.98). Results provided by calibration et al., 2013).
models were comparable in terms of accuracy to the conventional anal- On another study, the potential of near-infrared spectroscopy (NIR)
yses, revealing the feasibility for on-line or routine applications of NIR and multivariate analysis to replace reference methods in the character-
spectra regarding roasting process (Alessandrini et al., 2008). ization of some constituents of coffee and banana residues (leaves and
Another study focused on four of the most important constituents of husks) was investigated. The evaluated parameters were Klason lignin,
espresso and roasted coffee that are essential from a quality assurance acid soluble lignin, total lignin, extractives, moisture, ash and acid insol-
standpoint: total acidity, caffeine content, chlorogenic acids and roasted uble residue contents of 102 coffee residues samples. The precision of
bean colour. Similarly to the trend observed by Alessandrini et al. the NIR methodology performed well compared to the reference meth-
(2008), it was observed that water band intensity at 1450 (rst over- od for almost all the parameters, except for moisture. The models for all
tone of O\H stretching) and at 1940 nm (combination band of O\H the analytes exhibited R2 N 0.80, with the exception of ash and acid sol-
stretching and O\H deformation) decreased gradually when coffees uble lignin content, which were predicted poorly (R2 b 0.80). Thus, NIR
were roasted darker, since during roasting water reduction is propor- spectroscopy proved to be useful for chemical prediction of coffee resi-
tional to the degree of roasting. Other signicant spectral differences ap- dues, representing a faster and more economical alternative to the stan-
peared around 11001250, 14501700, 17801850, 19202000 and dard methodologies (Rambo, Amorim, & Ferreira, 2013).
20502200 nm, allowing the correlation of spectral variations with cer-
tain changes in chemical composition occurring during roasting, i.e., 4. Future trends
with the formation or degradation of certain compounds such as degra-
dation of chlorogenic acids, changes in carbohydrates and amino acids. The results of previous research works presented in this review con-
A comprehensive list of wavelength assignments for each attribute pre- rmed that IR spectral techniques are well suited for classication, au-
dicted is proposed by the authors. Thanks to the conjunction of pre- thentication and prediction of essential properties in coffee samples.
processing methods and multivariate calibration with NIR spectroscopy, Both FTIR and NIR spectroscopy have been successfully exploited. It
separate regression models were developed and could be used for the can work in tandem or even replace standard methods in situations
prediction of the referred quality parameters (Esteban-Diez et al., where classication, sorting, or identication of coffee varieties is re-
2004c). quired. NIRS calibrations allow performing coffee sample characterisa-
The feasibility of applying variable selection techniques to extract a tion in a short time comparatively to classical methods to discriminate
minimum number (maximum parsimony) of informative predictors the coffee varieties, which are laborious and require lengthy sample
from NIR spectra to measure in real time roasting colour and caffeine preparation. Considering the high spectral resolution used in recent
content was investigated to predict quality parameters of unknown cof- studies, deeper research is required to extract the useful information
fee samples. Several variable selection techniques were tested including and reject the voluminous data that do not contribute to the application.
iterative predictor weighting (IPW) (Forina et al., 1999), interactive As a result, it is expected that the spectroscopic techniques will over-
stepwise elimination (ISE) (Boggia, Forina, Fossa, & Mosti, 1997), and come other time-demanding methods in the near future for more com-
uninformative variable elimination (UVE) (Centner et al., 1996); and plex applications in monitoring coffee quality. Also, the abundant
approaches based on ordinary least-squares (OLS) regression, such as chemical features laid in the spectra will enable implementing some ap-
genetic algorithms coupled with OLS regression (GAOLS) (Leardi, plications such as authentication of superior coffee quality; detection of
Boggia, & Terrile, 1992; Lucasius & Kateman, 1991), and stepwise adulteration and defective grains and discrimination between different
orthogonalization of predictors (SELECT) (Forina, Lanteri, Casale, & coffee qualities. The accurate classication of different quality grades
Cerrato-Oliveros, 2007). Selected wavelengths related to O\H, C\H will be very important for correct pricing and categorization of coffee
and N\H stretching second and rst overtones mainly assigned to which provide some economical benets for producers by increasing
chlorogenic acids, carbohydrates, amino acids and water were observed consumer condence in the supplied products.
for modelling roasting colour and caffeine content. The application of The samples used in most of the studies reported were designed to
the stepwise orthogonalization of predictors provided notably im- cover as best as possible the large natural intrinsic variability of com-
proved regression models, with root-mean-square errors of the resid- mercially available coffees by considering different blends, roasting con-
uals in external prediction (RMSEP) equal to 3.68 and 1.46% for ditions and geographical origins. However, transfer of the technique out
roasting colour and caffeine content, respectively. The use in industrial of the laboratory will require its extension to commercial samples. The
scale of the calibration models proposed has the potential of reducing dynamic nature and particular needs of the coffee market would de-
analytical time, efforts, and costs of assessing these roasted coffee qual- mand a continuous and more exhaustive collection of calibration sam-
ity parameters, allowing for online monitoring of the coffee roasting ples that should be incessantly added to develop classication and
 D.F. Barbin et al. / Food Research International 61 (2014) 2332 31

prediction models using NIR spectroscopic techniques (Downey & principal component analysis. Journal of Agricultural and Food Chemistry, 45(1997),
46804686.
Boussion, 1996; Esteban-Dez et al., 2007). On the other hand, it must Blanco, M., & Villarroya, I., 2002. Trends in analytical chemistry, 21(4), 240250.
be emphasized that the application of NIR spectroscopy would not Boggia, R., Forina, M., Fossa, P., & Mosti, L., 1997. Chemometrics study and validation strat-
imply any additional effort regarding further analytical determinations egies in the structureactivity relationships of a new class of cardiotonic agents.
Quantitative StructureActivity Relationships, 16, 201206.
in future samples to be characterised. The measurement of reference Briandet, R., Kemsley, E. K., & Wilson, R. H., 1996. Approaches to adulteration detection in
chemical parameters is only required for samples used to build the cal- instant coffees using infrared spectroscopy and chemometrics. Journal of the Science
ibration dataset. Once calibration NIR spectra have been established, the of Food and Agriculture, 71, 359366.
Centner, V., Massart, D. L., de Noord, O. E., De Jong, S., Vandeginste, B.M., & Sterna, C., 1996.
pre-processing steps and prediction models obtained can be directly ap- Elimination of uninformative variables for multivariate calibration. Analytical Chemistry,
plied to future sample spectra without the need for measuring any ref- 68, 38513858.
erence values (Esteban-Dez et al., 2004a). In this way, from a practical Chen, L., Xue, X., Ye, Z., Zhou, J., Chen, F., & Zhao, J., 2011. Determination of Chinese honey
adulterated with high fructose corn syrup by near infrared spectroscopy. Food
application standpoint, the classication and prediction methodologies
Chemistry, 128(4), 11101114.
proposed only rely on NIR measurements. Chen, Q., Zhao, J., Zhang, H. And, & Wang, X., 2006. Feasibility study on qualitative and
IR spectroscopy techniques may nd applications in the coffee in- quantitative analysis in tea by near infrared spectroscopy with multivariate calibra-
dustry for simple product inspection, prediction of sensory features or tion. Analytica Chimica Acta, 572, 7784.
Cozzolino, D., Smyth, H. E., & Gishen, M., 2003. Feasibility study on the use of visible and
for quantication and prediction of sample composition. The non- near-infrared spectroscopy together with chemometrics to discriminate between
destructive and multivariate characteristics of NIR spectroscopy tech- commercial white wines of different varietal origins. Journal of Agricultural and Food
niques provide an interesting tool for quality monitoring changing the Chemistry, 51, 77037708.
Craig, A. P., Franca, A. S., & Oliveira, L. S., 2012a. Evaluation of the potential of FTIR and
measurements of coffee quality from the laboratories to the processing chemometrics for separation between defective and non-defective coffees. Food
lines. The use of FTIR and/or NIR spectroscopy for on/inline monitoring Chemistry, 132(2012), 13681374.
of quality in coffee industry is a challenge for the future. Craig, A. P., Franca, A. S., & Oliveira, L. S., 2012b. Discrimination between defective and
non-defective roasted coffees by diffuse reectance infrared Fourier transform spec-
troscopy. LWT Food Science and Technology, 47, 505511.
De Marchi, M., 2013. On-line prediction of beef quality traits using near infrared spectros-
5. Conclusion
copy. Meat Science, 94(4), 455460.
Downey, G., & Boussion, J., 1996. Authentication of coffee bean variety by near-infrared
This review has covered some of the recent applications of infrared reectance spectroscopy of dried extract. Journal of the Science of Food and
Agriculture, 71, 4149.
spectroscopy in coffee quality assessment. In aid of greater regulatory
Downey, G., Briandet, R., Wilson, R. H., & Kemsley, E. K., 1997. Near- and mid-infrared
inspection and consistent with the coffee industry's goal of providing spectroscopies in food authentication: Coffee varietal identication. Journal of
superior product quality, both FTIR and NIR spectroscopy have shown Agricultural and Food Chemistry, 45, 43574361.
the potential to enhance the knowledge and understanding of product Downey, G., Mcintyre, P., & Davies, A. N., 2003. Geographic classication of extra virgin
olive oils from the eastern Mediterranean by chemometric analysis of visible and
parameters along the production process. The main advantage is that near-infrared spectroscopy data. Applied Spectroscopy, 57, 158163.
it is a chemical-free assessment method where sample preparation Downey, G., Sheehan, E., Delahunty, C., O'Callaghan, D., Guinee, T., & Howard, V., 2005,
time is critically reduced or even eliminated, thus being a useful refer- JuneSeptemberr. Prediction of maturity and sensory attributes of Cheddar cheese
using near-infrared spectroscopy. International Dairy Journal, 15(69), 701709.
ence for traditional methods. The wide application of this system http://dx.doi.org/10.1016/j.idairyj.2004.06.013 (ISSN 09586946).
would seem to offer a number of potential advantages, including re- Ebrahimi-Najafabadi, H., Leardi, R., Oliveri, P., Casolino, M. C., Jalali-Heravi, M., & Lanteri,
duced labour costs, decreased errors during sample preparation or sub- S., 2012. Detection of addition of barley to coffee using near infrared spectroscopy
and chemometric techniques. Talanta, 99, 175179.
jective judgement, and the creation of product data in a real time for Esteban-Dez, I., Gonzlez-Siz, J. M., & Pizarro, C., 2004a. An evaluation of orthogonal sig-
documentation, traceability, and labelling. However, it is necessary for nal correction methods for the characterisation of arabica and robusta coffee varieties
researchers to further transfer the knowledge from laboratory scale to by NIRS. Analytica Chimica Acta, 514, 5767.
Esteban-Dez, I., Gonzlez-Siz, J. M., & Pizarro, C., 2004b. Prediction of roasting colour and
industrial lines, so that the coffee industry can realistically benet
other quality parameters of roasted coffee samples by near infrared spectroscopy. A
from the possibility of performing this non-destructive technique feasibility study. Journal of Near Infrared Spectroscopy, 12(5), 287297.
without additional laborious and time-consuming chemical analyses, Esteban-Dez, I., Gonzlez-Siz, J. M., & Pizarro, C. I., 2004c. Prediction of sensory proper-
ties of espresso from roasted coffee samples by near-infrared spectroscopy. Analytica
enabling early sorting of produce and thereby improving quality man-
Chimica Acta, 525, 171182.
agement. Considering the recent advances in multivariate modelling Esteban-Dez, I., Gonzlez-Siz, J. M., Saenz-Gonzalez, J. M., & Pizarro, C. I., 2007. Coffee va-
and the analytical requirements of the most recent concepts of quality, rietal differentiation based on near infrared spectroscopy. Talanta, 71, 221229.
it is anticipated that NIR techniques may progressively become a rou- Fagan, C. C., O'Donnell, C. P., O'Callaghan, D. J., Downey, G., Sheehan, E. M., Delahunty,
C. M., et al., 2007. Application of mid-infrared spectroscopy to the prediction of
tine method for coffee quality and safety assessment and process mon- maturity and sensory texture attributes of cheddar cheese. Journal of Food Science,
itoring and control. 72, E130E137.
Farah, A., Monteiro, M. C., Calado, V., Franca, A. S., & Trugo, L. C., 2006. Correlation between
cup quality and chemical attributes of Brazilian coffee. Food Chemistry, 98, 373380.
Acknowledgements Forina, M., Casolino, C., & Pizarro, C., 1999. Iterative predictor weighting (IPW) PLS: a
technique for the elimination of useless predictors in regression problems. Journal
of Chemometrics, 13, 165184.
The authors gratefully acknowledge the nancial support from the Forina, M., Lanteri, S., Casale, M., & Cerrato-Oliveros, M. C., 2007. Stepwise orthogonaliza-
Coordination for the Improvement of Higher Education Personnel tion of predictors in classication and regression techniques: An old technique
revisited. Chemometrics and Intelligent Laboratory Systems, 87, 252261.
(CAPES) strategic research initiative under the Brazilian Ministry of
Franca, A. S., Mendona, J. C. F., & Oliveira, S. D., 2005. Composition of green and roasted
Education, project number 23038.019085/2009-14. coffees of different cup qualities. LWT Food Science and Technology, 38, 709715.
Franca, A. S., Oliveira, L. S., Mendona, J. C. F., & Silva, X. A., 2005. Physical and chemical
attributes of defective crude and roasted coffee beans. Food Chemistry, 90, 8994.
References Garrigues, J. M., Bouhsain, Z., Garrigues, S., & de la Guardia, M., 2000. Fourier transform in-
frared determination of caffeine in roasted coffee samples. Journal of Analytical
Abbott, J. A., 1999. Quality measurement of fruits and vegetables. Postharvest Biology and Chemistry, 366(3), 319322.
Technology, 15(3), 207225. Gonzalez, A. G., Pablos, F., Martin, M. J., Leon-Camacho, M., & Valdenebro, M. S., 2001.
Abdel-Nour, N., Ngadi, M., Prasher, S., & Karimi, Y., 2011. Prediction of egg freshness and HPLC analysis of tocopherols and triglycerides in coffee and their use as authentica-
albumen quality using visible/near infrared spectroscopy. Food and Bioprocess tion parameters. Food Chemistry, 73(2001), 93101.
Technology, 4(5), 731736. He, J., Rodriguez-Saona, L. E., & Giusti, M. M., 2007. Mid infrared spectroscopy for juice au-
Alessandrini, L., Romani, S., Pinnavaia, G., & Rosa, M.D., 2008. Near infrared spectroscopy: An thentication Rapid differentiation of commercial juices. Journal of Agricultural and
analytical tool to predict coffee roasting degree. Analytica Chimica Acta, 625, 95102. Food Chemistry, 55, 44434452.
Bauer, R., Nieuwoudt, H., Bauer, F. F., Kossmann, J., Koch, K. R., & Esbensen, K. H., 2008. Huck, C. W., Guggenbichler, W., & Bonn, G. K., 2005. Analysis of caffeine, theobromine and
FTIR spectroscopy for grape and wine analysis. Analytical Chemistry, 80, 1371e1379. theophylline in coffee by near infrared spectroscopy (NIRS) compared to
Bicchi, C. P., Ombretta, M. P., Pellegrino, G., & Vanni, A.C., 1997. Characterization of roasted high-performance liquid chromatography (HPLC) coupled to mass spectrometry.
coffee and coffee beverages by solid phase microextractiongas chromatography and Analytica Chimica Acta, 538, 195203.
32 D.F. Barbin et al. / Food Research International 61 (2014) 2332

Huck, C. W., Maurer, R., Popp, M., Basener, N., & Bonn, G. K., 1999. Quantitative Fourier Perrone, D., Donangelo, C. M., & Farah, A., 2008. Fast simultaneous analysis of caffeine,
transform near infrared reectance spectroscopy (NIRS) compared to high perfor- trigonelline, nicotinic acid and sucrose in coffee by liquid chromatographymass
mance liquid chromatography (HPLC) of a avone in Flos Primulae veris extractives. spectrometry. Food Chemistry, 110, 10301035.
Pharmaceutical and Pharmacological Letters, 9, 2629. Pizarro, C., Esteban-Diez, I., & Gonzlez-Siz, J. -M., 2007. Mixture resolution according to
Illy, A., & Viani, R., 1996. Espresso coffee: The chemistry of quality. London: Academic Press. the percentage of robusta variety in order to detect adulteration in roasted coffee by
Inn, F. A., Garrigues, S. And, & Guardia, M., 2006. Combination of mid- and near-infrared near infrared spectroscopy. Analytica Chimica Acta, 585, 266276.
spectroscopy for the determination of the quality properties of beers. Analytica Pizarro, C., Esteban-Diez, I., Gonzalez-Saiz, J. -M., & Forina, M., 2007. Use of near-infrared
Chimica Acta, 571, 167174. spectroscopy and feature selection techniques for predicting the caffeine content and
Isaksson, T., & Naes, T., 1988. The effect of multiplicative scatter correction (MSC) and lin- roasting color in roasted coffees. Journal of Agricultural and Food Chemistry, 55,
earity improvement in NIR spectroscopy. Applied Spectroscopy, 42, 12731284. 74777488.
ISO International Standard, 1978. Green coffee Determination of moisture (routine meth- Pizarro, C., Esteban-Diez, I., Nistal, A. -J., & Gonzlez-Siz, J. -M., 2004. Inuence of data
od). ISO 1447:1978. pre-processing on the quantitative determination of the ash content and lipids in
ISO International Standard, 1983. Green coffee Determination of loss in mass at 105 C. roasted coffee by near infrared spectroscopy. Analytica Chimica Acta, 509, 217227.
ISO 6673:1983. Rambo, M. K. D., Amorim, E. P., & Ferreira, M. M. C., 2013. Potential of visible-near infrared
ISO International Standard, 2001. Green coffee Determination of moisture (basic reference spectroscopy combined with chemometrics for analysis of some constituents of cof-
method). ISO 1446:2001. fee and banana residues. Analytica Chimica Acta, 775, 4149.
Jing, M., Cai, W., & Shao, X., 2010. Quantitative determination of the components in corn Reh, C. T., Gerber, A., Prodolliet, J., & Vuataz, G., 2006. Water content determination in green
and tobacco samples by using near-infrared spectroscopy and multiblock partial least coffee Method comparison to study specicity and accuracy. Food Chemistry, 96,
squares. Analytical Letters, 43(12), 19101921. 423430.
Kemsley, E. K., Ruault, S., & Wilson, R. H., 1995. Discrimination between Coffea arabica and Ribeiro, J. S., Ferreira, M. M. C., & Salva, T. J. G., 2011. Chemometric models for the quan-
Coffea canephora variant robusta beans using infrared spectroscopy. Food Chemistry, titative descriptive sensory analysis of Arabica coffee beverages using near infrared
54, 321326. spectroscopy. Talanta, 83(2011), 13521358.
Leardi, R., Boggia, R., & Terrile, M., 1992. Genetic algorithms as a strategy for feature selec- Ribeiro, J. S., Salva, T. J., & Ferreira, M. M., 2010. Chemometric studies for quality control of
tion. Journal of Chemometrics, 6, 267282. processed Brazilian coffees using DRIFTS. Journal of Food Quality, 33, 212227.
Liao, Y., Fan, Y., & Cheng, F., 2012. On-line prediction of pH values in fresh pork using Santos, K. M., Moura, M. F. V., Azevedo, F. G., Lima, K. M. G., Raimundo, I. M., Jr., & Pasquini,
visible/near-infrared spectroscopy with wavelet de-noising and variable selection C., 2012. Classication of Brazilian coffee using near-infrared spectroscopy and mul-
methods. Journal of Food Engineering, 109(4), 668675. tivariate calibration. Analytical Letters, 45(7), 774781.
Lucasius, C. B., & Kateman, G., 1991. Genetic algorithms for large-scale optimization in Santos, J. R., Sarragua, M. C., Rangel, A. O. S. S., & Lopes, J. A., 2012. Evaluation of green
chemometrics: An application. Trends in Analytical Chemistry, 10, 254261. coffee beans quality using near infrared spectroscopy: A quantitative approach.
Lyman, D. J., Benck, R., Dell, S., Merle, S., & Murray-Wijelath, J., 2003. FTIR-ATR analysis of Food Chemistry, 135, 18281835.
brewed coffee: Effect of roasting conditions. Journal of Agricultural and Food Schulz, H., 2004. Application in analysis of coffee, tea, cocoa, tobacco, spices, medicinal
Chemistry, 51, 32683272. and aromatic plants, and related products. In C. Roberts, J. Workman, & J. Reeves
Mendona, J. C. F., Franca, A. S., & Oliveira, L. S., 2009. Physical characterization of (Eds.), Near-infrared spectroscopy in agriculture. Agronomy Monograph, No. 44.
non-defective and defective Arabica and Robusta coffees before and after roasting. (pp. 345376). Madison, USA: American Society of Agronomy/Crop Science of
Journal of Food Engineering, 92, 474479. America/Soil Science Society of America.
Mendona, J. C. F., Franca, A. S., Oliveira, L. S., & Nunes, M., 2008. Chemical characterization Telo, R. F., Martins, J. P. A. And, & Ferreira, M. M. C., 2009. Sorting variables by using in-
of non-defective and defective green Arabica and Robusta coffees by Electrospray formative vectors as a strategy for featuring selection in multivariate regression.
IonizationMass Spectrometry (ESIMS). Food Chemistry, 111, 490497. Journal of Chemometrics, 23, 3241.
Morgano, M.A., Faria, C. G., Ferro, M. F., Bragagnolo, N., & Ferreira, M. M. C., 2008. Wang, J., Jun, S., Bittenbender, H. C., Gautz, L., & Li, Q. X., 2009. Fourier transform infrared
Determinao de umidade em caf cru usando espectroscopia NIR e regresso spectroscopy for Kona coffee authentication. Journal of Food Science, 74(5), C385C391.
multivariada [Determination of moisture in raw coffee by near infra-red reectance Wang, D., Min, S. -G., Duan, J., Xiong, Y. -M., & Li, Q. -Q., 2012. Simultaneous determination
spectroscopy and multivariate regression]. Cincia e Tecnologia de Alimentos, 28(1), of instant coffee, plant fat and sugar content in liquid coffee beverage by diffuse reec-
1217. tance near-infrared spectroscopy. Guang Pu Xue Yu Guang Pu Fen Xi/Spectroscopy and
Myles, A. J., Zimmerman, T. A., & Brown, S. D., 2006. Transfer of multivariate classication Spectral Analysis, 32(4), 982984.
models between laboratory and process near-infrared spectrometers for the discrim- Weyer, L. G., 1985. Near-infrared spectroscopy of organic substances. Applied Spectroscopy
ination of green Arabica and Robusta coffee beans. Applied Spectroscopy, 60(10), Reviews, 21, 143.
246A272A (10971216. 1198-1203(6)). Woodcock, T., Fagan, C. C., O'Donnell, C. P., & Downey, G., 2008. Application of near and
Osborne, B. G., Fearn, T., & Hindle, P. H., 1993. Practical NIR spectroscopy with applications mid-infrared spectroscopy to determine cheese quality and authenticity. Food &
in food and beverage analysis (2nd ed.). Harlow, UK: Longman Scientic and Bioprocess Technology, 1(2), 117129. http://dx.doi.org/10.1007/s11947-007-0033-y.
Technical. Xie, L., Ying, Y., Ying, T., Yu, H. And, & Fu, X., 2007. Discrimination of transgenic tomatoes
Pscoa, R. N. M. J., Magalhes, L. M., & Lopes, J. A., 2013. FT-NIR spectroscopy as a tool for based on visible/near-infrared spectra. Analytica Chimica Acta, 584, 379384.
valorization of spent coffee grounds: Application to assessment of antioxidant prop- Yan-De, L., Yi-Bin, Y., Xiaping, F. And, & Guisan, L., 2007. Experiments on predicting sugar
erties. Food Research International, 51(2013), 579586. content in apples by FT-NIR technique. Journal of Food Engineering, 80, 986989.
Pasquini, C., 2003. Near Infrared Spectroscopy: Fundamentals, practical aspects and ana- Zhang, X., Li, W., Yin, B., Chen, W., Kelly, D. P., Wang, X., et al., 2013. Improvement of near
lytical applications. Journal of Brazilian Chemical Society, 14(2), 198219. infrared spectroscopic (NIRS) analysis of caffeine in roasted Arabica coffee by vari-
Pedro, A.M. K., & Ferreira, M. M. C., 2005. Nondestructive determination of solids and ca- able selection method of stability competitive adaptive reweighted sampling
rotenoids in tomato products by near-infrared spectroscopy and multivariate calibra- (SCARS). Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 114,
tion. Analytical Chemistry, 77, 25052511. 350356.