S. Saranwong, J. Sornsrivichai and S. Kawano, J. Near Infrared Spectrosc.

11, 175–181 (2003) 175

Performance of Portable NIR Instrument for Brix Determination of Intact Mango

S. Saranwong, J. Sornsrivichai and S. Kawano, J. Near Infrared Spectrosc. 11, 175–181 (2003)

Performance of a portable near infrared

instrument for Brix value determination of
intact mango fruit
Sirinnapa Saranwong,a,c Jinda Sornsrivichaia and Sumio Kawanob
a
Department of Biology, Faculty of Science, Chiang Mai University, Chiangmai 50202, Thailand
b
National Food Research Institute, 2-1-12 Kannondai, Tsukuba 305-8642, Japan

The performance of a commercially-available, portable NIR instrument “FT20” for fruit quality evaluation was in-
vestigated using mango fruits. The calibration result for the FT20 was compared with the result for the Foss
NIRSystems 6500. Partial least squares (PLS) regression was used to make the calibration equations. It was found
that the calibration equation using spectral data obtained with the FT20 had similar accuracy to that using Foss
NIRSystems 6500 spectra. The SEPs of calibration equations were 0.40 °Brix for both instruments.

Keywords: near infrared (NIR) spectroscopy, non-destructive quality evaluation, portable NIR instrument, partial
least squares (PLS) regression, Brix value, mango

Introduction NIR spectra of these fruits.1–6 Even if these instru-

The internal quality of fruit is important to con-
sumer satisfaction. Sweetness is one of the major
factors that affect taste in many kinds of fruits.
Therefore, consumer understanding and apprecia-
tion of sweetness is essential to the marketing of
fresh fruits. Timely, objective measurements of Brix
value (an index of soluble solids in fresh fruits)
would help consumers to appreciate the fresh prod-
ucts they purchase.
Near infrared (NIR) spectroscopy has been
adopted for Brix determination of fruits such as
peaches, apples and oranges in many countries for
almost a decade. Usually, NIR instruments for
research or sorting machines are used to measure
c Present
address: National Food Research Institute, 2-1-12
Kannondai, Tsukuba 305-8642, Japan
ments are highly precise, their application to field
research, such as monitoring of chemical changes of
developing fruits on trees, are limited by their large
size and weight.
Presently, there are several commercially-
available portable NIR instruments offered for de-
termining Brix value. Ishibashi et al.7 reported that
the accuracy of the Kubota NIR instrument “Fruit
Selector” for Brix determination of intact tomatoes
was 0.33 °Brix in SEP. Temma et al.8 also showed
that the portable NIR instrument developed by their
research centre had excellent potential in determin-
ing Brix value of intact apples. However, the accu-
racy of a portable NIR instrument when compared to
a research one is still not clear.
This research was conducted to compare the per-
formance of a commercially-available portable NIR
instrument with that of a research-type
spectrophotometer.

© NIR Publications 2003, ISSN 0967-0335

176 Performance of Portable NIR Instrument for Brix Determination of Intact Mango

Materials and methods

Materials
A total of 95 Philippine mango fruits (Mangifera
indica cv. Caraboa) which were commercially avail-
able were used as samples in this experiment. The
fruits, packed in corrugated fibreboard boxes, were
purchased at the wholesale market, transported to
our laboratory at the National Food Research Insti-
tute (NFRI) and then kept in the cold storage room at
5°C for the experiment on the following day. Six
hours before the acquisition of spectral data, all
fruits were moved from storage to the instrument
laboratory (25°C ambient).

Spectral acquisition
NIR spectra of mango fruits were acquired with
the portable NIR instrument “Fruit Tester 20” 28 mm
(FANTEC, Kosai-city, Japan) and the research type
NIR instrument “Model 6500” (Foss NIRSystems,
Silver Spring, USA), hereafter referred to as the
“FT20” and “NIRS6500”, respectively. The spectral 9.3 mm
range of the FT20 was 600–1000 nm while that of 14 mm
25.4 mm
the NIRS6500 was 400–1100 nm. For comparison,
certain characteristics and measuring conditions of
the two instruments are given in Table 1 and Table 2, (a) (b)
respectively.
Figure 1. Sample presentation of mango fruit for FT20
Both the FT20 and the NIRS6500 utilised inter- instrument (a) and NIRS6500 instrument (b). The area
actance probes. The FT20 used a probe in which the filled with crossed lines is the illuminator, while the area
illuminator portion and detector portion were side filled with dots is the detector.

Table 1. Specifications of FT20 and NIRS6500 instruments.

Items FT20 NIRS6500

Wavelength range 600–1000 nm 400–1100 nm
Detector Silicon diode array Silicon
a
Measuring time 3–5 s 30 s
Size of measuring head 28 mm 25.4 mm
Distance between illuminator and detector 14 mm 9.3 mm
Instrument dimension 25(W) × 20(H) × 10 cm(D) 25(W) × 35(H) × 29 cm(D)
Instrument weight 2.8 kg 25 kg
a Measuring time is not included standard measurement
S. Saranwong, J. Sornsrivichai and S. Kawano, J. Near Infrared Spectrosc. 11, 175–181 (2003) 177

Table 2. NIR measuring conditions of the FT20 and NIRS6500 instruments.

Instruments NIR measuring conditions

Items Value
FT20 Integral time 30 ms
Average 32 scans
Wavelength measured 600–1000 nm
Reference material Ceramic plate
Reference measurement Every sample
NIRS6500 Time per scan 50 ms
Average 50 scans
Wavelength measured 400–1100 nm
Reference material Teflon sphere
Reference measurement Every six samples

by side [Figure 1(a)]. The NIRS6500 used a probe Chemical analysis

with a concentric outer ring illuminator and an inner A portion of the flesh of each fruit which was illu-
ring detector [Figure 1(b)]. The illumination area of minated by NIR (about 5 cm diameter and 10 mm
the FT20 was 14 mm while that of the NIRS6500 deep) was taken and analysed for Brix value with a
was 18.6 mm according to its ring-type geometry. digital refractometer (Model PR101, ATAGO, To-
For the FT20, the spectral acquisition was per- kyo, Japan). Brix value for each fruit used to develop
formed under normal room lightning (i.e. no light- the calibrations was the average of duplicate deter-
tight box used) as shown in Figure 1(a). Each sample minations. Statistical data pertaining to the Brix val-
spectrum was the average of 32 scans. The reference ues are given in Table 3.
measurement of a ceramic plate inside the measur-
ing head was carried out automatically before each
sample spectrum was acquired.
For the NIRS6500, the measurement procedures Table 3. Characteristics of calibration and validation
followed those in previous research.2 Each sample sample sets of mango used.
spectrum was the average of 50 scans. The reference
measurement was performed with a Teflon sphere Items Calibration set Validation set
(8 cm diameter) every six fruits. Na 50 45
Interactance measurements with both instru-
ments were taken from the fruit shoulder (90° from Range 12.3–19.9 12.5–19.3
the fruit stem axis). The temperature of each fruit Mean 16.4 16.3
was fixed by dipping it in a water bath maintained at
25°C for a period of 10 min just prior to scanning. A SD 1.80 1.59
thin film of polyethylene was used to separate the
Unit °Brix °Brix
fruit from the water during temperature equilibra-
tion. aN is the number of samples
178 Performance of Portable NIR Instrument for Brix Determination of Intact Mango

0.60 1.20
(a) Original spectra (a) Original spectra M
0.40 1.00 H
L

0.20 0.80

log (1/R)
log (1/R)

0.00 0.60
M
-0.20 H
0.40
L
-0.40 0.20
-0.60 0.00
750 800 850 900 950 1000 750 800 850 900 950 1000

0.15 0.15
(b) Second derivative spectra
0.10 (b) Second derivative spectra
0.10
d2log (1/R)

d2log (1/R)
0.05 0.05
0.00 0.00
-0.05 -0.05
-0.10 844 978 -0.10 840
962 962
-0.15 -0.15
750 800 850 900 950 1000 750 800 850 900 950 1000
Wavelength (nm) Wavelength (nm)
Figure 2. Original spectra (a) and second derivative spec- Figure 3. Original spectra (a) and second derivative spec-
tra (b) of typical mango fruits having low (L), medium (M) tra (b) of typical mango fruits having low (L), medium (M)
and high (H) Brix values measured with FT20 instrument. and high (H) Brix values measured with NIRS6500 instru-
ment.

and the NIRS6500 are shown in Figures 2 and 3, re-

Data analysis
The Fruits20® program (FANTEC, Kosai-city,
Japan) was used for spectral data transformation
from the FT20. The Near Infrared Spectral Analysis
Software (NSAS®) program (Foss NIRSystems, Sil-
ver Spring, USA) was used for second derivative
pretreatment. The parameters for the second deriva-
tive reported in our previous work6 (segment = 4 nm;
gap = 4 nm) were used for the NIRS6500 spectra and
segment = 8 nm and gap = 8 nm for the FT20 spectra.
The Unscrambler® program (CAMO, Oslo, Norway)
was used for partial least squares (PLS) regression.
spectively. The log (1/R) values of each fruit seemed
to be affected by scattering condition of sample
rather than Brix content. The difference in shape of
the original spectra acquired by the FT20 and the
NIRS6500 was due primarily to the use of different
reference materials (the ceramic plate for the FT20
and the Teflon sphere for the NIRS6500). Note that,
after transformation, the second derivative spectra
for the two instruments were similar and that the
peaks of the water band appeared at 844 nm and
962 nm for the FT20, and 840 nm and 962 nm for the
NIRS6500.

Calibration and validation

Results and discussion
NIR spectra of intact mango
Original [log (1/R)] and second derivative [d2log
(1/R)] spectra of typical intact mangoes having low,
medium and high Brix values measured by the FT20
PLS calculations were carried out using second
derivative spectra for the FT20 and the NIRS6500.
The calibration and validation results for the FT20
are shown in Table 4. Table 4 shows that the best
wavelength region for the FT20 spectra was from
S. Saranwong, J. Sornsrivichai and S. Kawano, J. Near Infrared Spectrosc. 11, 175–181 (2003) 179

Table 4. Calibration and validation results for Brix value using spectral data measured with FT20 in the wavelength region
from 900 to 1000 nm and from 850 to 1000 nm.

Calibration Wavelength region Fa R SEC SEP Bias RPD

A 900–1000 nm 6 0.94 0.64 0.55 0.13 2.89
B 850–1000 nm 5 0.98 0.39 0.40 0.02 3.98
a F is the number of factors used in the calibration equation

R: Multiple correlation coefficients

SEC: Standard error of calibration
SEP: Bias-corrected standard error of prediction
Bias: The average of difference between actual value and NIR value
RPD: Ratio of standard deviation of reference data in validation set to SEP

850 to 1000 nm, which was wider than the result 200
Regression coefficient (K )

(900–1000 nm) reported in our previous work.6 This 150 (a) 900 ~ 1000nm
is discussed below. 100
In order to explain the result from PLS regression, 984
50
the regression coefficient plots of the FT20 spectra 0
using the wavelength region from 900 to 1000 nm,
-50
and from 850 to 1000 nm are shown in Figure 4(a)
-100
and Figure 4(b), respectively. Figure 4(a) shows that 913 951
-150
the wavelength of 978 nm had an important role in 978
-200
calibration A in Table 4 as was the case for our previ-
ous result using the NIRS6500 with the same wave- 850 900 950 1000
length region (900–1000 nm).6 However, due to the 80
Regression coefficient (K )

second derivative treatment and the spectral limita- 60 (b) 850 ~ 1000nm
tion of 1000 nm for the FT20, there was no signal 40
984
from 978 to 1000 nm as shown in Figure 2(b). There- 20
0 878
fore, using spectral data from 978 to 984 nm as indi-
cated in Figure 4(a) was considered to cause errors -20
giving high SEP value. On the other hand, Figure -40 978
4(b), the regression coefficient plot for calibration -60 954
913
B, shows that the wavelength of 878 nm had an im- -80
portant role while the peak at 978 nm became 850 900 950 1000
smaller than that in calibration A. These plots indi- Wavelength (nm)
cated that using spectral data at 878 nm in place of
978 nm could compensate for the missing data at Figure 4. Regression coefficient plots for PLS calibration
of mango spectra measured with the FT20 over the wave-
978 nm and improve the accuracy of the FT20 cali- length region of 900–1000 nm (a) and 850–1000 nm (b).
bration equation.
The calibration and validation results for the
FT20 and the NIRS6500 are shown in Table 5. Scat-
ter plots between actual and predicted Brix values of
the validation sample set are also shown in Figure 5. to 1000 nm would have similar accuracy to the
The result showed that the calibration equation us- NIRS6500 equation. Both equations had SEPs of
ing FT20 spectra in the wavelength region from 850 0.40 °Brix. Using the 95% pair t-test, there were no
180 Performance of Portable NIR Instrument for Brix Determination of Intact Mango

Table 5. Calibration and validation results for Brix value using spectral data measured with FT20 and NIRS6500 instru-
ments.

Instrument Wavelength region Fa R SEC SEP Bias RPD

FT20 850–1000 nm 5 0.98 0.39 0.40 0.02 3.98
NIRS6500 850–1000 nm 5 0.97 0.43 0.40 0.02 3.98
a F is the number of factors used in the calibration equation

R: Multiple correlation coefficients

SEC: Standard error of calibration
SEP: Bias-corrected standard error of prediction
Bias: The average of difference between actual value and NIR value
RPD: Ratio of standard deviation of reference data in validation set to SEP

Conclusion
The calibration equations developed using spec-
tra measured with the portable NIR instrument
“FT20” and the research NIR instrument
“NIRS6500” had the same accuracy in predicting
Brix value of intact mango. With its compact size,
fast scan, low cost and sufficient accuracy, it was
concluded that the portable NIR instrument “FT20”
was the promising instrument for non-destructive
quality evaluation of fruits such as mango, espe-
cially on trees.

Acknowledgement
This research was supported by the Royal Golden
Jubilee PhD Program under the Thailand Research
Fund (TRF).

References

1. M. Iwamoto, S. Kawano and Y. Ozaki, J. Near

Figure 5. Scatter plots for predicting Brix value of mango
in the validation sample set using spectra measured with
the FT20 (a) and the NIRS6500 (b).
significant differences between the actual and the
NIR-predicted values in both the FT20 and the
NIRS6500 equations.
Infrared Spectrosc. 3, 179 (1995).
2. S. Kawano, H. Watanabe and M. Iwamoto, J. Ja-
pan. Soc. Hort. Sci. 61(2), 445 (1992).
3. S. Kawano, T. Fujiwara and M. Iwamoto, J. Ja-
pan. Soc. Hort. Sci. 62(2), 465 (1993).
4. S. Kawano, H. Abe and M. Iwamoto, J. Near In-
frared Spectrosc. 3, 211 (1995).
5. K. Miyamoto, M. Kawauchi and T. Fukuda, J.
Near Infrared Spectrosc. 6, 267 (1998).
S. Saranwong, J. Sornsrivichai and S. Kawano, J. Near Infrared Spectrosc. 11, 175–181 (2003) 181

6. S. Saranwong, J. Sornsrivichai and S. Kawano, 8. T. Temma, K. Hanamatsu and F. Shinoki, J.

J. Near Infrared Spectrosc. 9, 287 (2001). Near Infrared Spectrosc. 10, 77 (2002).
7. H. Ishibashi, T. Takada, Y. Suzuki, S.
Morimoto, M. Kashu and R. Yamauti, Proc. 15th Received: 18 June 2002
Symp. on Nondestructive Techniques, Tsukuba, Accepted: 15 October 2002
Japan, p.125 (1999). Web Publication: 8 July 2003