Food Chemistry 132 (2012) 1368–1374

Contents lists available at SciVerse ScienceDirect

Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem

Evaluation of the potential of FTIR and chemometrics for separation between

defective and non-defective coffees
Ana Paula Craig a, Adriana S. Franca a,b,⇑, Leandro S. Oliveira a,b
a
Programa de Pós-Graduação em Ciência de Alimentos/UFMG, Av. Antônio Carlos, 6627, 31270-901 Belo Horizonte, MG, Brazil
b
Departamento de Engenharia Mecânica/UFMG, Av. Antônio Carlos, 6627, 31270-901 Belo Horizonte, MG, Brazil

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

Article history: The objective of this work was to evaluate the potential of Fourier transform infrared spectroscopy (FTIR)
Received 18 April 2011 for the discrimination of defective and non-defective coffee beans. Defective (black, immature and sour)
Received in revised form 14 October 2011 and non-defective Arabica coffee beans were submitted to FTIR analysis by transmittance readings
Accepted 29 November 2011
employing KBr discs and reflectance readings employing attenuated total reflectance (ATR) and diffuse
Available online 6 December 2011
reflectance (DR) accessories. Multivariate statistical analysis (PCA, clusters) was performed in order to
verify the possibility of discrimination between defective and non-defective coffee samples. A clear sep-
Keywords:
aration between defective and non-defective coffee beans was observed, based on both PCA and cluster
Coffee
FTIR
analysis of the reflectance spectra (ATR and DR accessories) and of the first derivatives of the transmit-
DRIFTS tance spectra (KBr discs). Such results indicate that FTIR analysis has the potential for the development
ATR of a fast and reliable analytical methodology for the discrimination between defective and non-defective
coffee beans.
Ó 2011 Elsevier Ltd. Open access under the Elsevier OA license.

1. Introduction Brazilian internal market. Thus, the roasting industry in Brazil has
been using these defective beans in blends with healthy ones, and
The presence of defective coffee beans depreciates the quality of overall, a low-grade roasted coffee is consumed in the country (Oli-
the coffee beverage consumed worldwide (Mancha Agresti, Franca, veira, Franca, Mendonça, & Barros-Junior, 2006).
Oliveira, & Augusti, 2008). The intrinsic defects (sour, black and Colour sorting is the major procedure employed for separation
immature beans) are the ones that, when roasted, contribute the of defective and non-defective coffee beans prior to roasting. In
most to the depreciation of the coffee beverage quality. According Brazil, manual sorting is usually employed for bean quality classi-
to Clarke and Macrae (1987), black beans are usually associated fication and electronic sorting is employed in farms and coopera-
with a heavy flavour, sour beans contribute to sour and oniony tives of producers for the actual removal of defective beans. In
tastes, while immature beans will impart astringency to the bever- the electronic sorters, coffee beans pass, one by one, by an elec-
age. The negative effect that such beans have on coffee quality can tronic eye or camera system, and depending on wavelength mea-
be associated with specific problems that occur during harvesting surements, the bean is either allowed to pass or it is shot with a
and post-harvest processing operations. Black beans result from puff of air into a reject pile (Franca & Oliveira, 2008). However,
dead beans within the coffee cherries or from beans that fall natu- such procedure is not efficient for the separation of sour and
rally on the ground by action of rain or over-ripening (Mazzafera, immature beans. Actually, in order make sure that such defects
1999). The presence of sour beans can be associated with ‘overfer- are effectively removed from a specific coffee lot, colour sorting
mentation’ during wet processing and with improper drying or machines are usually set up to allow non-defective coffees to be
picking of overripe cherries, whereas immature beans come from also removed if their colour is similar to that of sour or immature
immature fruits (Clarke & Macrae, 1987; Mendonça, Franca, beans. As a consequence of this, the coffee lots that are rejected as
Oliveira, & Nunes, 2008). Defective beans represent about 20% of defective may present a high percentage of good coffee, as pointed
the total coffee produced in Brazil and, although they are separated out in studies employing machine sorted mixtures or low quality
from the non-defective beans prior to commercialisation in exter- Arabica coffees from different origins and crops (Farah, Monteiro,
nal markets, the majority of these beans are dumped on the Calado, Franca, & Trugo, 2006; Franca, Mendonça, & Oliveira,
2005; Franca, Oliveira, Mendonça, & Silva, 2005; Vasconcelos, Fran-
⇑ Corresponding author at: Programa de Pós-Graduação em Ciência de Alimentos/ ca, Glória, & Mendonça, 2007). The same problem is present if sep-
UFMG, Av. Antônio Carlos, 6627, 31270-901 Belo Horizonte, MG, Brazil. Tel.: +55 31 aration by sieving is employed (Franca, Oliveira, et al., 2005;
34093512; fax: +55 31 34433783. Mendonça, Franca, & Oliveira, 2009).
E-mail addresses: [email protected], [email protected] (A.S. Franca).

0308-8146 Ó 2011 Elsevier Ltd. Open access under the Elsevier OA license.
doi:10.1016/j.foodchem.2011.11.121
 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374 1369

Recent studies have shown that some chemical parameters ZnSe cell was employed. Although the ATR-FTIR technique has been
could be employed for the separation between defective and mostly applied for analysis of liquid samples, there are recent stud-
non-defective green coffee beans of a given variety (Arabica or ies that employ ATR for direct readings on solid food products (e.g.
Robusta). Examples include levels of histamine, determined by cheese, meats), given that it requires minimal sample preparation
high performance liquid chromatography – HPLC (Vasconcelos and variations in sample thickness have been shown not to affect
et al., 2007) and electrospray-ionisation mass spectrometry (ESI- the intensity of the bands (Argyri, Panagou, Tarantilis, Polysiou, &
MS) profiles (Mendonça et al., 2008). However, most of the em- Nychas, 2010; Koca, Rodriguez-Saona, Harper, & Alvarez, 2007). In
ployed instrumental techniques and analytical procedures are time order to obtain a constant sample mass, a small metal recipient
demanding, expensive and involve a considerable amount of man- 2.4 mm thick and presenting an aperture of the same size of the
ual work. Recent studies have also shown that FTIR-based meth- ATR accessory (79 mm long and 10 mm wide) was placed over the
ods, in combination with chemometric techniques, can be ZnSe ATR crystal. The ground coffee sample (2 g, particle diameter
successfully applied in the food industry, in association with food <0.39 mm) was then placed inside the metal recipient and pressed
quality evaluation (Rodriguez-Saona & Allendorf, 2011). FTIR- with a spatula in order to obtain the best possible contact with the
based methods are fast, reliable, simple to perform and do not crystal. The empty recipient was used to obtain the background
require sample pre-treatment. Such technique provides simple spectrum. The approximate total times required for sample prepara-
and reproducible means of handling food products with nonde- tion were 40 min (transmittance readings), 20 min (DR readings)
structive analyses, with the sampling/analysis procedure usually and 5 min (ATR readings). Regardless of the sample preparation pro-
taking only a few minutes. cedure, all spectra were recorded within a range of 4000–700 cm1
There are a few studies that have focused on FTIR applied to cof- with a 4 cm1 resolution and 20 scans. Spectra treatment consisted
fee analysis, employing either roasted coffee or aqueous extracts of background subtraction, baseline correction and normalisation.
(e.g. coffee beverage). The specific applications were discrimina- The statistical package XLSTAT Sensory 2010 (Addinsoft, New York)
tion between Arabica and Robusta varieties (Kemsley, Ruault, & was employed for the chemometric calculations.
Wilson, 1995), detection of glucose, starch or chicory as adulter-
ants of freeze-dried instant coffees (Briandet, Kemsley, & Wilson,
3. Results and discussion
1996), evaluation of roasting conditions (Lyman, Benck, Dell,
Merle, & Murray-Wijelath, 2003), geographical discrimination
The average values of measured colour parameters for non-
(Wang, Jun, Bittenbender, Gautz, & Li, 2009) and separation be-
defective and defective coffee samples are shown in Table 1. The
tween decaffeinated and regular roasted coffees (Ribeiro, Salva, &
measurements were based on the CIE Lab three dimensional
Ferreira, 2010). Thus, the objective of this work was to evaluate
cartesian (xyz) colour space, represented by: Luminosity (L⁄), rang-
the potential of Fourier transform infrared (FTIR) spectroscopy in
ing from 0 (black) to 100 (white) – z axis; parameter a⁄, represent-
the characterisation and discrimination between defective and
ing the green–red colour component – x axis; and parameter b⁄,
non-defective coffee beans prior to roasting.
representing the blue–yellow component – y axis. However, chro-
maticity can be better represented and discussed in terms of polar
coordinates, so a⁄ and b⁄ values were converted to chroma (c⁄) and
2. Materials and methods
hue angle (h), since these parameters can be directly associated to
colour intensity or saturation (c⁄) and to colour tone (h):
Arabica green coffee samples, acquired from Café Fino Grão
(Contagem, MG), were comprised of coffee beans obtained from c ¼ ½a2 þ b 1=2
2
ð1Þ
different cooperatives located in Minas Gerais State, Brazil, that
were rejected by colour sorting machines. Black, sour (separated 
h ¼ tan1 ½b =a  ð2Þ
into light and dark coloured), immature and non-defective beans
were manually picked to constitute separate sampling lots. Colour The results presented in Table 1 representing measurements
measurements were performed using a tristimulus colorimeter performed on whole coffee beans, i.e., evaluation of the bean sur-
(HunterLab Colorflex 45/0 Spectrophotometer, Hunter Laborato- face colour, show that black and dark sour beans presented lower
ries, VA, USA), with standard illumination D65, and colorimetric luminosity values than non-defective, immature and light sour
normal observer angle of 10°, employing both whole and ground ones, indicating that this parameter can be successfully employed
coffee samples. Colour measurements were performed thrice for only to separate black and dark sour defects prior to roasting. Such
each sample. results are in agreement with previous studies on physical attri-
A Shimadzu IRAffinity-1 FTIR Spectrophotometer (Shimadzu, butes of defective coffee beans (Mendonça et al., 2009). It can also
Japan) with a DLATGS (Deuterated Triglycine Sulphate Doped with be observed that non-defective, immature and black beans pre-
L-Alanine) detector was used in the measurements that were all per- sented higher values of hue angle in association with a greenish
formed in a dry atmosphere at room temperature (20 ± 0.5 °C). For tone, whereas the lower values of hue angle observed for sour
the transmittance readings, ground coffee samples (particle diame- beans are associated to a yellowish brown tone. Black and dark
ter <0.5 mm) were mixed with potassium bromide at a 1/50 ratio (w/ sour beans presented the lowest values of colour saturation. Colour
w). This mixture (0.1 g) was then compressed into a thin KBr disc un- measurements taken for ground samples represent an average col-
der a pressure of 7845 kPa for 5 min. The spectrum of a clean KBr disc our of the material, taking into account both the surface and inte-
(without coffee) was used for subtraction (background spectrum). rior. Luminosity values were higher for ground beans compared to
For the reflectance readings, both diffuse and attenuated modes whole ones, as a consequence of the fact that the bean surface is
were employed. Diffuse reflectance (DR) measurements were per- darker than its core. Values for colour parameters for both whole
formed in diffuse reflection mode with a Shimadzu diffuse reflec- and ground samples were similar to those reported on previous
tance sampling accessory (DRS8000A). The ground coffee sample studies employing defective coffees from different crops and ori-
(1 mg, particle diameter <0.15 mm) was mixed with KBr (100 mg) gins (Franca, Oliveira, et al., 2005; Mendonça et al., 2009).
and then 23 mg of this mixture was placed inside the sample port. The results presented in Table 1 for whole beans indicate that
Pure KBr was employed as reference material (background spec- the monochromatic colour separation procedure commonly
trum). For the attenuated reflectance measurements (ATR-FTIR), a employed in farms and cooperatives will only be effective in the
horizontal ATR sampling accessory (ATR-8200HA) equipped with case of black and dark sour defects. This can be corroborated by
1370 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374

Table 1
Average colour attributes of non-defective and defective coffee samples.

Whole beans Ground beans

⁄ ⁄ ⁄ ⁄
Sample L a b h c L⁄ a⁄ b⁄ h c⁄
a c b b a b c
Non-defective 46.1 ± 1.1 2.7 ± 0.3 19.0 ± 0.8 81.8 ± 1.0 19.2 ± 0.8 59.6 ± 1.3 2.3 ± 0.6 19.4 ± 0.6b 83.3 ± 1.5c 19.6 ± 0.6b
Immature 43.1 ± 1.1b 1.8 ± 0.3d 20.3 ± 0.8a 84.9 ± 0.6a 20.4 ± 0.8b 61.2 ± 1.1a 0.5 ± 0.2d 21.7 ± 0.9a 88.7 ± 0.4b 21.7 ± 0.9a
Sour (light) 37.1 ± 2.1c 6.5 ± 1.0a 16.8 ± 0.9c 68.8 ± 3.4c 18.1 ± 0.8c 55.7 ± 0.6c 4.1 ± 0.3a 18.1 ± ± 0.5c 77.3 ± 0.7e 18.6 ± 0.6c
Sour (dark) 29.6 ± 2.0d 3.7 ± 0.4b 10.2 ± 1.4d 70.0 ± 1.8c 10.9 ± 1.4d 48.2 ± 0.5d 3.5 ± ± 0.1b 18.0 ± 0.2c 78.9 ± 0.4d 18.4 ± 0.2c
Black 27.6 ± 1.3d 0.9 ± 0.2e 6.7 ± 0.8e 82.4 ± 1.4b 6.8 ± 0.8e 36.0 ± 0.9e –1.3 ± 0.4e 10.9 ± 0.9d 97.0 ± 2.1a 11.0 ± 0.9d

Average ± standard deviation. Average values followed by the same letter in the same column do not differ significantly by the Tukey test at 5% probability.

the score plots obtained for PCA analysis of colour parameters of ter one being correlated with the asymmetric stretching of CAH
whole beans (see Fig. 1). Data matrices for PCA analysis were bonds of methyl (ACH3) group in the caffeine molecule and the
assembled so that each row corresponded to a sample and each peak region being successfully used to develop predictive models
column to a colour parameter. The first two principal components for quantitative analysis of caffeine (Paradkar & Irudayaraj,
(PCs) explained 66% and 32% of the data variance, respectively. 2002). The sharp band at 1743 cm1 has been also observed on
Four distinct groups can be perceived, separated by quadrant: light FTIR studies of roasted coffee (Kemsley et al., 1995; Lyman et al.,
sour (positive PC1, positive PC2); dark sour (negative PC1, positive 2003; Wang et al., 2009). Kemsley et al. (1995) reported that a
PC2); black (negative PC1, negative PC2); and non-defective/imma- band at 1744 cm1 was larger in Arabica in comparison to Robusta
ture (positive PC1, negative PC2). The first component allowed for sample and attributed this to the carbonyl (C@O) vibration associ-
the separation between darker and lighter samples, being mostly ated to the ester group in triglycerides. The study by Lyman et al.
affected by luminosity values. Separation by the second compo- (2003) also associated the bands in that region to aliphatic esters
nent can be associated to black, immature and non-defective beans (1755–1740 cm1). A band at 1658 cm1 appears in the spectra ob-
presenting a greenish tone as opposed to the yellowish hue of sour tained by KBr transmission, as can be seen in Fig. 2a, and it is also
beans. Such results indicate that even sorting systems that employ associated to caffeine absorption (Lyman et al., 2003). Ribeiro et al.
bi-chromatic light measurements will not be able to completely (2010) reported that wavenumbers in the range of 1700–
separate immature and non-defective coffee beans. 1600 cm1 are highly related to chlorogenic acids and caffeine con-
Typical FTIR spectra obtained for green coffee samples are centration in coffees. We confirmed the identification of the bands
shown in Fig. 2. A full assignment of the spectral bands is quite previously associated to caffeine (2922, 2852 and 1658 cm1) by
challenging problem and is not the scope of this work. Further- FTIR-ATR analysis of aqueous extracts of non-defective coffee
more, FTIR literature data on coffee is only available for roasted spiked with caffeine. There was a significant increase in peak
samples, so a direct comparison cannot be done. Nonetheless, a intensity with the increase in caffeine concentration (spectra not
few qualitative aspects of the spectra can be discussed. The spectra shown).
obtained by transmission and reflectance are similar from a quali- Other bands that appear at lower intensity can be viewed in the
tative point of view, in the sense that the most significant bands range of 1600–1000 cm1. According to Kemsley et al. (1995),
can be viewed in both types of spectrum. Also, higher intensity Briandet et al. (1996), and Lyman et al. (2003), chlorogenic acids
of peaks can be observed in the spectra that employed KBr (trans- present strong absorption in the region of 1300–1150 cm1. Chlor-
mission and diffuse reflectance, Fig. 2a and b, respectively), in the ogenic acids correspond to a large family of esters formed between
1800–800 cm1 range. The two sharp bands that can be viewed in quinic acid and one to four residues of certain trans-cinnamic acids,
the 3000–2800 cm1 range (2924–2922 and 2852 cm1) have also most commonly caffeic, p-coumaric and ferulic (Clifford, Kirkpa-
been reported for both Arabica and Robusta roasted coffee sam- trick, Kuhnert, Roozendaal, & Salgado, 2008). Axial CAO deforma-
ples, but no identification was attempted (Kemsley et al., 1995). tion of the quinic acid occurs in the range of 1085–1050 cm1
Nonetheless, studies of FTIR analysis of caffeine on soft drinks have while OAH angular deformation occurs between 1420 and
also reported two sharp peaks at 2882 and 2829 cm1, with the la- 1330 cm1. The CAOAC ester bond also absorbs in the 1300–
1000 cm1 range (Silverstein, Webster, & Kiemle, 2005). Thus, the
bands at 1381–1376, 1161–1153 and 1053 cm1 could be associ-
ated to chlorogenic acids. The wavenumber range of 1400–
3 900 cm1 is characterised by vibrations of several types of bonds,
including CAH, CAO, CAN and PAO (Sablinskas, Steiner, & Hof,
2 2003; Wang et al., 2009). Carbohydrates have been previously
shown to exhibit several absorption bands in this region (Briandet
et al., 1996; Kemsley et al., 1995), so it is expected that this class of
PC2 (32.4%)

1 compounds will also contribute to the several bands appearing in

this region.
PCA analysis of the KBr transmission spectra, employing (a)
0
baseline correction and normalisation and (b) first derivatives is
displayed in Fig. 3. The analysis was based on a 24  1192 data ma-
-1 trix assembled so that each row corresponded to a sample and
each column represented the spectra data at a given wavelength.
For these specific analyses, dark and light coloured sour beans
-2
-3 -2 -1 0 1 2 3 were grouped together, because the sampling preparation required
PC1 (66.3%) a reasonable amount of time. In the case of PCA based on the spec-
tra (Fig. 3a), the two first components accounted for 80.2% of the
Fig. 1. PCA scores scatter plot of L⁄a⁄b⁄ colour parameters for whole coffee beans (PC1 total sample variance. A certain amount of sample separation can
vs. PC2). s, non-defective; , immature; , sour (light); N, sour (dark); d, black. be observed, mainly for non-defective (negative PC1, positive
 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374 1371

1654 (a) 40
2923
1 (a)
30
1053
1743 1153
1378 0.8 20

10

PC2 (31.7%)
Normalized absorbance
2852 0.6
0

0.4 -10

-20
0. 2
1268 -30

-40
0 -60 -40 -20 0 20 40 60
3200 2700 2200 1700 1200 700
PC1 (48.6%)
-0.2
wavenumber (cm-1 ) 40
(b)
1
30
2924 1743 1654 (b)
1381 1161 20
1053 0.8
10

PC2 (21.6%)
2852
0
0.6
Normalized absorbance

-10
0.4
1273 -20
1541
-30
0.2
-40
-40 -30 -20 -10 0 10 20 30 40
0 PC1 (31.2%)
3200 2700 2200 1700 1200 700
Fig. 3. PCA scores scatter plot (PC1 vs. PC2) of KBr transmission FTIR spectra
-0.2
submitted to (a) normalisation and baseline correction and (b) first derivatives. (s,
non-defective; , immature; , sour (light and dark mixed); d, black).

wavenumber (cm-1 ) -0.4

tively. In this case, sample scattering diminished considerably
2922 and there was a clear separation between non-defective and defec-
(c) 1 tive coffee beans. Three separate groups can be identified: (i) non-
defective, (ii) immature and (iii) black/sour (fermented). Evalua-
tion of the PC1 and PC2 loadings plots (not shown) did not indicate
0.8
2852 specific regions of the spectra that could be directly associated to
1743 group separation.
Normalized absorbance

0.6 PCA analysis of the DR spectra, employing baseline correction

1458 and normalisation is displayed in Fig. 4a. Analysis was based on a
45  1188 data matrix assembled so that each row corresponded
1376 0.4
1153 to a sample and each column represented the spectrum data at a gi-
1534
ven wavelength. The two first components accounted for 94.2% of
1044 0.2 the total sample variance. A clear separation between defective
and non-defective coffees can be observed, with four major groups:
non-defective (positive PC1, negative PC2), immature/light sour
0
3200 2700 2200 1700 1200 700 (positive PC1, positive PC2), dark sour (negative PC1, positive
PC2), and black (negative PC1, negative PC2). Clustering of imma-
wavenumber (cm-1 ) -0.2 ture and light sour defects was also observed by Mendonça et al.
(2008) for analysis of the ESI(+)-MS profiles. Evaluation of the load-
Fig. 2. Typical green coffee bean FTIR spectra obtained by (a) transmittance
ings plot (not shown) indicated that the spectral ranges that pre-
readings employing KBr discs and reflectance readings employing (b) DR and (c)
ATR accessories (spectra manipulation consisted of baseline correction and
sented the highest influence on sample grouping were 2980–2850
normalisation). and 1560–800 cm1 corresponding to immature/light sour sam-
ples, 1700–1570 cm1 corresponding to non-defective beans,
3100–3000, 1980–1760 in reference to dark sour, and 2000–
PC2), immature (negative PC1, negative PC2), and fermented (sour/ 1985 cm1 corresponding to black beans. Group separation was
black positive PC1), but there was a significant scattering of the not enhanced by taking derivatives of the spectra.
samples. In the case of PCA based on the first-derivative of the PCA analysis of the ATR reflectance spectra, employing baseline
spectra (Fig. 3b), the first and second principal components ac- correction and normalisation is displayed in Fig. 4b. Analysis was
counted for 31.2% and 21.6% of the total sample variance, respec- based on a 54  1188 data matrix assembled so that each row cor-
1372 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374

40 ters: non-defective, immature and fermented (black/sour). Diffuse

(a) reflectance spectra (DR) also provided complete separation be-
30
tween non-defective and defective coffee beans, as can be seen in
20 Fig. 5b. Four major clusters can be viewed, in reference to black,
dark sour, non-defective and immature/light sour. In this case over-
10 all clustering can be also related to sample surface colour, with the
PC2 (32.4%)

darker samples (black and dark sour) being completely separated

0
from the remaining lighter samples. ATR spectra did not provide a
-10 complete separation between defective and non-defective coffees,
as shown in Fig. 5c. Three major clusters can be viewed, the first
-20 one comprised of non-defective and light sour coffees and the other
two containing immature, black and dark sour coffees. Grouping of
-30
black and dark sour coffees has been previously reported for analy-
-40 sis of ESI(+)-MS profiles (Mendonça et al., 2008), in association with
-60 -40 -20 0 20 40 60
lower sucrose levels of such beans in comparison to non-defective
PC1 (61.8%)
ones. This may also be the case here, since reduction in sucrose lev-
els will probably occur after fermentation. Also, in the case of
30 immature beans, low sucrose levels can be associated to the bean
(b)
20 maturity state whereas for black and sour beans, sucrose reduction
is probably due to fermentation. Grouping of black and immature
10 beans was also observed in the analysis of the volatile profile of
roasted coffees (Mancha Agresti et al., 2008), in association to the
0
PC2 (19.6%)

occurrence of black beans being associated to the fermentation of

-10 immature ones. The same study also reported grouping of non-
defective and sour beans (no separation between dark and light
-20 sour), in association to sour beans corresponding to non-defective
coffees that underwent fermentation during handling and process-
-30
ing after harvest.
-40 The results obtained in the present study showed that FTIR-
based methods seem an attractive alternative for developing a fast
-50
-60 -40 -20 0 20 40 60 80 routine method for discrimination of non-defective and defective
PC1 (66.7%) coffees. Derivatives of the spectra based on transmittance readings
employing KBr discs allowed grouping according to the specific
Fig. 4. PCA scores scatter plot (PC1 vs. PC2) of reflectance spectra submitted to type of defect. However, we foresee two major disadvantages with
normalisation and baseline correction (a) DRIFTS and (b) ATR-FTIR (s, non- this particular sampling technique. The first one is related to the
defective; , immature; , sour (light); N, sour (dark); d, black).
time and care required during sample preparation. The second
responded to a sample and each column represented the spectra one is related to the small amount of coffee (0.002 g) that is em-
data at a given wavelength. The two first components accounted ployed for each analysis. DRIFTS also provided a clear separation
for 86.3% of the total sample variance. The first component provided between defective and non-defective coffees, with the advantage
separation of the evaluated samples into two major groups: non- of less sample preparation, i.e., ground coffee is just mixed with
defective/light sour (positive PC1) and black/dark sour/immature KBr and directly placed in the DR accessory to be analysed, without
(negative PC1). Evaluation of the loadings plot (not shown) indi- the further need of pressing and preparation of a clear disc
cated that the spectral ranges that presented the highest influence (20 min). Nonetheless, the amount of coffee employed for analy-
on PC1 values in association with the black/dark sour/immature sis is even smaller (0.00023 g). Although the previous sampling
group were the following: 1554–1482, 1797–1776 and 3100– methods provided satisfactory separation between defective and
3020 cm1. The only significant band that can be observed in the non-defective coffees, the fact that the sample mass is quite small
ATR spectra (Fig. 2b) in those ranges is the one at 1534 cm1. It could be a problem if quantitative analysis is needed, i.e., deter-
was also present in the spectra obtained by Lyman et al. (2003) mining the amount of defective beans that are present in a given
for aqueous extracts of roasted coffee, regardless of roasting condi- coffee sample. It is noteworthy to point out that, although ATR-
tions, although no identification was attempted. We herein infer FTIR did not provide complete separation between non-defective
that this band is most likely associated to C@C stretch in a nitrogen and light sour coffees, such results can still be deemed satisfactory,
based ring, such as caffeine or trigonelline, both present in signifi- given that the main problem with colour sorting is the separation
cant amounts in raw and roasted coffees. In the case of PCA based of immature and non-defective beans. Thus, ATR-FTIR could be
on the first-derivative of the spectra, the first and second principal viewed as a complementary procedure to be employed after colour
components accounted for 28.6% and 12.6% of the total sample var- sorting and, among the evaluated sampling techniques, it is the
iance, respectively. No group separation could be observed. one that allows the use of larger samples (2 g), and thus will prob-
To verify the similarity between the samples and to single out ably be more appropriate for quantitative evaluation.
some classes, agglomerative hierarchical clustering (AHC) was ap- The method used by coffee traders for coffee classification is
plied to the set of variables employed for PCA. Ward’s hierarchical based on the types and amount of defective beans present in a
clustering method and chord distance were employed to establish sample (usually 300 g out of a bag of about 60 kg). The professional
clusters and calculate dissimilarity coefficients, respectively. The sorter is trained to identify the defective and non-defective beans
resulting dendrograms are displayed in Fig. 5. Group clustering con- solely by their appearance, with colour being the most effective
firmed the PC analysis. In the case of transmittance spectra ob- characteristic contributing for the differentiation (others would
tained with KBr discs (Fig. 5a), a clear separation between non- be size, shape, etc.). The methodology herein studied, employing
defective and defective coffees can be observed, in three major clus- the ground beans, could be used to replace the professional sorters
 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374 1373

1.2
(a)
1.0

0.8

Dissimilarity
0.6

0.4

0.2

0.0
ND
ND
ND
ND
ND
ND

BL
BL
BL

BL
SO
SO
BL
BL
SO
SO
SO
SO
IM
IM
IM
IM
IM
IM
1.6
(b)
1.4

1.2

1.0
Dissimilarity

0.8

0.6

0.4

0.2

0.0
ND
ND
ND
ND
ND
ND
ND
ND
ND
BL
BL
BL
BL
BL
BL
BL
BL
BL
DS
DS
DS
DS
DS
DS
DS
DS
DS

IM
IM
IM
IM
IM
IM
IM
IM
IM
LS
LS
LS
LS
LS
LS
LS
LS
LS

6
(c)
5

4
Dissimilarity

0
ND

ND
ND
ND
ND
ND
ND
ND
ND
ND
DS
BL
DS
BL
BL

BL
BL
BL
BL
BL
DS
DS
DS
DS
BL
BL
BL
DS
DS
BL
DS
IM

IM
IM
IM
IM
IM
IM
IM
IM

IM

IM
LS
LS
LS
LS
LS
LS
LS
LS
LS
LS
LS

Fig. 5. Hierarchical cluster analysis (HCA) of (a) first derivatives of FTIR spectra obtained by transmittance readings employing KBr discs and normalised FTIR spectra
obtained by reflectance readings employing (b) DR and (c) ATR accessories (ND, non-defective; SO, sour; LS, light sour; DS, dark sour; IM, immature; BL, black).

in the process of classification of coffee samples for commercialisa- readings employing KBr discs and DRIFTS readings. PCA and AHC
tion, thus eliminating the subjectivity of the current procedure. results based on normalised ATR-FTIR reflectance spectra indicated
the separation of the samples into two major groups: non
-defective/light sour and black/dark sour/immature. The results
4. Conclusions obtained in the present study confirm that FTIR analysis presents
the potential for the development of an analytical methodology
The feasibility of employing FTIR as a methodology for the sep- for the discrimination between defective and non-defective coffee
aration between defective and non-defective coffees was evaluated beans. Further studies will be conducted employing larger sets of
and successfully demonstrated. PCA and AHC results indicated that samples in order to develop predictive models. The methodology
non-defective and defective coffee samples could be separated into will be also tested for roasted coffee samples.
distinct groups, based on transmittance or reflectance spectra It is noteworthy to point out, however, that FTIR-based method-
obtained by mixing the coffee samples with KBr, i.e., transmittance ologies are devised for dealing with particles, liquids or solids of
1374 A.P. Craig et al. / Food Chemistry 132 (2012) 1368–1374

large smooth surfaces, making them inappropriate for use with Franca, A. S., Oliveira, L. S., Mendonça, J. C. F., & Silva, X. A. (2005). Physical and
chemical attributes of defective crude and roasted coffee beans. Food Chemistry,
whole coffee beans. Thus, the methodology proposed herein does
90, 89–94.
not allow for the actual separation of the single defective beans Franca, A. S., Mendonça, J. C. F., & Oliveira, S. D. (2005). Composition of green and
in an automated production processes, but represents a first step roasted coffees of different cup qualities. LWT – food science and technology, 38,
towards its achievement, in the sense that infrared spectral ranges 709–715.
Franca, A. S., & Oliveira, L. S. (2008). Chemistry of defective coffee beans. In E. N.
that presented the highest influence on group separation were Koeffer (Ed.), Food chemistry research developments (pp. 105–138). New York,
identified. When the beans are ground, the chemical makeup of NY, USA: Nova Publishers.
the bean surface will remain embedded in the sample and thus will Kemsley, E. K., Ruault, S., & Wilson, R. H. (1995). Discrimination between Coffea
arabica and Coffea canephora variant robusta beans using infrared spectroscopy.
contribute to the makeup of the FTIR spectra. Also, with the excep- Food Chemistry, 54, 321–326.
tion of the lipid fraction, which presents a somewhat non uniform Koca, N., Rodriguez-Saona, L. E., Harper, W. J., & Alvarez, V. B. (2007). Application of
distribution within the beans (the wax fraction is only found in the Fourier transform infrared spectroscopy for monitoring short-chain free fatty
acids in Swiss cheese. Journal of Dairy Science, 90, 3596–3603.
outer layer of the bean), all the other classes of compounds are Lyman, D. J., Benck, R., Dell, S., Merle, S., & Murray-Wijelath, J. (2003). FTIR-ATR
evenly distributed throughout the bean, including its surface. Thus, analysis of brewed coffee: Effect of roasting conditions. Journal of Agricultural
one should expect the spectra for the whole beans to be similar to and Food Chemistry, 51, 3268–3272.
Mancha Agresti, P. C. M., Franca, A. S., Oliveira, L. S., & Augusti, R. (2008).
those for the ground beans. Considering the suitability of the meth- Discrimination between defective and non-defective Brazilian coffee beans by
od for ground samples shown in this study, the proposal for future their volatile profile. Food Chemistry, 106, 787–796.
implementation in automatic sorting of single defective beans Mazzafera, P. (1999). Chemical composition of defective coffee beans. Food
Chemistry, 64, 547–554.
would be to employ infrared radiation at specific wavelengths, in
Mendonça, J. C. F., Franca, A. S., Oliveira, L. S., & Nunes, M. (2008). Chemical
a way similar to that of the colour sorting machines. characterization of non-defective and defective green Arabica and Robusta
coffees by electrospray ionization-mass spectrometry (ESI-MS). Food Chemistry,
111, 490–497.
Acknowledgements Mendonça, J. C. F., Franca, A. S., & Oliveira, L. S. (2009). Physical characterization of
non-defective and defective Arabica and Robusta coffees before and after
roasting. Journal of Food Engineering, 92, 474–479.
The authors acknowledge financial support from the following Oliveira, L. S., Franca, A. S., Mendonça, J. C. F., & Barros-Junior, M. C. (2006).
Brazilian Government Agencies: CNPq and FAPEMIG. Proximate composition and fatty acids profile of green and roasted defective
coffee beans. LWT – Food Science and Technology, 39, 235–239.
Paradkar, M. M., & Irudayaraj, J. (2002). Rapid determination of caffeine content in
References soft drinks using FTIR-ATR spectroscopy. Food Chemistry, 78, 261–266.
Ribeiro, J. S., Salva, T. J., & Ferreira, M. M. (2010). Chemometric studies for quality
Argyri, A. A., Panagou, E. Z., Tarantilis, P. A., Polysiou, M., & Nychas, G.-J. E. (2010). control of processed Brazilian coffees using DRIFTS. Journal of Food Quality, 33,
Rapid qualitative and quantitative detection of beef fillets spoilage based on 212–227.
Fourier transform infrared spectroscopy data and artificial neural networks. Rodriguez-Saona, L., & Allendorf, M. E. (2011). Use of FTIR for rapid authentication
Sensors and Actuators B: Chemical, 145, 146–154. and detection of adulteration of food. Annual Review of Food Science and
Briandet, R., Kemsley, E. K., & Wilson, R. H. (1996). Approaches to adulteration Technology, 2, 467–483.
detection in instant coffees using infrared spectroscopy and chemometrics. Sablinskas, V., Steiner, G., & Hof, M. (2003). Applications. In G. Gauglitz & T. Vo-Dinh
Journal of the Science of Food and Agriculture, 71, 359–366. (Eds.), Handbook of spectroscopy (pp. 89–168). Weinheim, Germany: Wiley-VCH.
Clarke, R. J., & Macrae, R. (1987). Coffee: Technology (Vol. 2). London: Elsevier Silverstein, M. R., Webster, F. X., & Kiemle, D. (2005). Spectrometric identification of
Applied Science. organic compounds. Hoboken, NJ, USA: Wiley.
Clifford, M. N., Kirkpatrick, J., Kuhnert, N., Roozendaal, H., & Salgado, P. R. (2008). Vasconcelos, A. L. S., Franca, A. S., Glória, M. B. A., & Mendonça, J. C. F. (2007). A
LC–MSn analysis of the cis isomers of chlorogenic acids. Food Chemistry, 106, comparative study of chemical attributes and levels of amines in defective
379–385. green and roasted coffee beans. Food Chemistry, 101, 26–32.
Farah, A., Monteiro, M. C., Calado, V., Franca, A. S., & Trugo, L. C. (2006). Correlation Wang, J., Jun, S., Bittenbender, H. C., Gautz, L., & Li, Q. X. (2009). Fourier transform
between cup quality and chemical attributes of Brazilian coffee. Food Chemistry, infrared spectroscopy for Kona coffee authentication. Journal of Food Science, 74,
98, 373–380. C385–C391.