Table 1 The number of measuring sessions weekly per cultivar, year and orchard
Year Orchard
Jonagold Elstar
Boskoop Golden Delicious
Cox’s Orange Pippin Gala
Braeburn 7
7 1997
7 Velm
6 5
– –
8 7
6 7
Velm 6
1998 5
8 Rillaar
1998 8
6 7
7 –
– 4
2 2
2 2
– –
1998 –
Lendelede
The objective of this study was two-fold, i to evaluate the potential to measure the internal
quality characteristics of apples by VISNIR-spec- troscopy in the preclimacteric phase; and ii to
predict the optimal harvest date based on VIS NIR spectra. A special attempt was made to
establish sufficiently robust calibration models that are applicable to all the different cultivars
over different years.
2. Materials and methods
2
.
1
. Fruit The experiment was performed over 2 years
1997 and 1998. The apples were picked during a maximum of 8 weeks before the commercial pick-
ing date at the experimental stations ‘Nationale Proeftuin voor Grootfruit’, in Velm, Belgium, and
‘Fruitteeltcentrum’ in Rillaar, Belgium; and at a commercial orchard in Lendelede, Belgium. These
orchards
are located
in scattered
locations throughout Belgium. Because of this, the soil
types differ from sandy loam Lendelede and Ril- laar to loam Velm. This set-up was chosen in
order to create robust models without orchard or year effects. The experimental schedule is sum-
marised for each orchard, cultivar and season in Table 1. For every measurement session, eight
apples per cultivar were randomly picked at eye level, in the middle of the canopy of eight differ-
ent trees. The apples were transported to the laboratory and stored under ambient conditions
prior to measurement. All measurements were carried out on the same day or the day after
picking. In total 952 apples were collected.
2
.
2
. Quality parameters The firmness was measured with a Magness –
Taylor penetrometer with an 11-mm diameter plunger. The juice that was released by the firm-
ness test was used to measure the soluble solids with a digital refractometer Atago Co. Ltd.,
Tokyo, Japan. The acidity was determined through titration of 10 ml apple juice mixture in
groups of two or three apples with 0.1 N NaOH to a pH of 8.8 phenolphthalein.
2
.
3
. Collection of VISNIR spectra From each apple, four reflection spectra 380 –
2000 nm, wavelength increment 0.5 nm were taken at four equidistant positions along the
equator with a spectrophotometer Optical Spec- trum Analyser OSA 6602, Rees Instruments
Ltd., Goldalming, UK in a 045° configuration Fig. 1.
The bundled detecting fibres and the bundled source fibres were placed on a black holder type
6151 under an angle of 45°. The light source Dual light source, type 6290 consisted of a 12
V100 W tungsten halogen lamp type Philips
Fig. 1. Set-up of VISNIR equipment.
7724.M28. This source is usable in the visible and infrared region. The reflected light is captured
by a grating monochromator. The spectrum 380 – 2000 nm was collected into two parts. A
Si-detector type 6611, 380 – 1080 nm and an InGaAs-detector type 6614, 1080 – 2000 nm each
measure one part of the whole spectrum and those two spectra are connected at 1080 nm. Every
spectrum was divided by a reference spectrum taken from a BaSO
4
-plate to minimise the light source ageing. Each measured reflection spectrum
was the average of five individual optical scans from 380 to 2000 by 0.5 nm increments Rees
Instruments, Macam Photometrics Ltd., UK. Lammertyn et al. 2000 found that NIR light
with a wavelength between 380 and 2000 nm penetrated into the apple flesh for up to 1 cm
depending on the wavelength.
2
.
4
. Data analysis The averaged reflection spectra were analysed
with the statistical program for multivariate cali- bration The Unscrambler CAMO AS, Trond-
heim, Norway. The spectrum was pre-processed first by calculating a three points segment moving
average. After normalisation of the data, the number of measuring points was reduced 14 times
7 nm increments. The second derivative spectra were calculated using the method of Savitzky and
Golay 1964 to correct for additive and multi- plicative effects in the spectra Martens and Naes,
1987. The pre-processed data were used in the statistical analysis together with the quality
parameters. The technique used was PLS. This is a projection method, like PCR, but uses both the
independent and dependent variables to find the regression model with its PLS components. PLS
often needs fewer latent variables to reach the optimal solution because the focus is on the de-
pendent variables De Jong, 1993. Cross valida- tion in groups of 20 samples was used to
validate the models. Extreme outliers were re- moved from the data set. The accuracy of the
calibration and validation are defined by SEC and SEP, as follows:
SEC = 1
I
c
− 1
I
c
i = 1
yˆ
i
− y
i 2
1 SEP =
1 I
p
− 1
I
p
i = 1
yˆ
i
− y
i
− bias
2
2 with yˆ
i
, the predicted value of the ith observation; y
i
, the measured value of the ith observation; I
c
, the number of observations in the calibration set;
I
p
, the number of observations in the validation set and bias = 1I
p I
p
i = 1
yˆ
i
− y
i
.
3. Results