Post on 06-Oct-2020
Application of NIRS to Olives,
Olive Oil, and Cheese
Products
Emil W. Ciurczak Castelo-Branco, Portugal
28 May 2015
emil@ciurczak.com
Doramaxx Consulting
Why Near-Infrared?
Used first by US Dept. Agriculture in 1960s for wheat, soy,
corn, fruits; later cotton, milk, tobacco, etc.
Can be used for “as is” plants and fruits a well as
processed materials, such as cheese and milk
Can be used in the field as well as laboratory
Minimum sample preparation, no chemicals or dilutions
Once calibrated, may be used by anyone in any situation
Cost of equipment and training is quickly offset by major
time savings
Doramaxx Consulting
Determination of Chlorophyll
in Olive Oil
There are various grades of olive oil;
extra virgin regular light
Extra Virgin olive oil is considered the highest quality, the first pressing
from freshly prepared olives. (It has a greenish-yellow tint and a
distinctively fruity aroma because of the high levels of chlorophyll
and other volatile materials extracted from the fruit.
Regular olive oil is collected with the help of a warm water slurry to
increase yield. (It is pale yellow in color, with a slight aroma,
because it contains fewer volatile compounds)
Light olive oil is very light in color and has no aroma because it has
been processed under pressure to remove the chlorophyll and
volatile compounds. (It is commonly used for frying because it
does not affect the taste of fried foods and it is relatively
inexpensive).
Doramaxx Consulting
Non-destructive assessment of olive fruit
ripening by a portable NIR
A portable NIR was used for the determination of oil and moisture
contents in intact olives.
In this test, spectral data were recorded in the region from 1100 to 2300
nm at 1 nm intervals under two different experimental conditions:
on-tree in the field in Trial 1
under laboratory room conditions in Trial 2.
Calibration models were developed and evaluated using PLS
regression separately for each trial set and for the combined
group of samples.
Doramaxx Consulting
Non-destructive assessment of olive fruit
ripening by a portable NIR
The combined model showed predictive statistics within the range of the
individual models (R = 0.89 and RMSECV = 1.99 for oil content
and R = 0.88 and RMSECV = 2.06 for moisture content),
considered acceptable as an increase in the model robustness
could be expected.
These results encourage the use of portable NIR spectroscopy to
monitor olive fruit ripening and to decide the optimal harvesting
date on the basis of oil and moisture content.
Although slightly better results were obtained under laboratory room
conditions, the results obtained on-tree in the field were also
accurate enough to determine the optimal harvest date of each
cultivar.
Doramaxx Consulting
Experimental Method
In trial 1, NIR spectra of fruits on-tree were obtained, samples were
then taken to laboratory.
In trial 2, fruits were harvested, and brought to laboratory spectra
collected.
The average spectra of five fruits per date were used for later analyses
in both trials. Spectra were acquired in absorbance with a portable
spectrophotometer, between 1100 to 2300 nm at 1 nm intervals.
Each spectrum is the average of 50 spectra acquired on the fruit
equator with continuous measuring for a total scanning time of 5
sec. Spectra collection was controlled with a laptop computer
(SNAP32 software)
Doramaxx Consulting
Methodology
After spectral collection, fruit samples were processed in the laboratory
for analysis by reference methods.
Fresh samples were weighed and then dried in an oven (105°C for 42h)
to determine moisture content.
The oil content of dried samples was recorded by NMR Minispec
NMS100 (Bruker).
Calibration models were developed separately for each trial set and for
the combined group of samples.
Full cross-validation (i.e. leaving-one-out) was used to determine the performance of
the models and no outliers were removed in any step of the calibration process.
Correlation between actual and predicted constituent values (r) and standard error of
cross validation (RMSECV) were used to test the performance of calibrations
(Shenk and Westerhaus, 1995).
Doramaxx Consulting
Results and Discussion Reference data and Spectral Features
Environmental conditions and ripening stages have provided wide
ranges of variability for the characteristics evaluated (Table 1).
For the combined group oil content ranged from 4.80 to 29.84 % and
moisture content from 48.12 to 69.57 %.
In both trials 1 and 2 oil content increases at the beginning of the
experiment and then stabilizes or even slightly decreases at the
end of the ripening period (Figure 1).
Moisture content showed the opposite trend decreasing during the
ripening period; both oil and moisture content were highly
correlated (R = –0.61, p < 0.001).
Different patterns were observed in trials 1 and 2 probably due to the
different climatic conditions of each experimental area.
Doramaxx Consulting
The average raw spectra and coefficient of variation of olive fruit
samples from T1 and T2 are shown in Figure 2: two water bands
~1460 nm & 1950 nm and oil around 1210 & 1730 nm.
Calibration development for each trial set
Calibration models were developed for moisture and oil contents for
each trial independently and for the combined data set.
The number of PLS factors, correlation coefficient and predictive error
(RMSECV and RER) obtained for models are shown in Table 2.
R values for oil content and moisture were slightly lower and RMSECV
was higher for model 1, due to different conditions during data
acquisition.
On-tree fruit spectral data were used in model 1, while the spectral data
of model 2 were obtained under constant laboratory room
conditions.
Doramaxx Consulting
Results and Discussion Reference data and spectral features
The selection of different cultivars, environmental conditions, and
ripening stages have provided wide ranges of variability for the
characteristics evaluated (Table 1).
In both trials 1 and 2 oil content increases at the beginning of the
experiment and then stabilizes or even slightly decreases at the
end of the ripening period (Figure 1).
Moisture content showed the opposite trend decreasing during the
ripening period.
Oil and moisture content, were highly correlated (r= –0.61, p < 0.001).
Different patterns were observed in trials 1 and 2, probably due to the
different climatic conditions of each experimental area.
Doramaxx Consulting
The average raw spectra and COV of samples from T1 and T2 are
seen in Figure 2.
Spectra are characterized by two water bands ~1460 nm and 1950 nm
and oil ~1210 nm and 1730 nm.
Calibration development for each trial set
Calibration models were developed for moisture and oil contents for
each trial independently and for the combined data set.
The number of PLS factors, correlation coefficient and predictive error
obtained for models are shown in Table 2.
R values for oil content and moisture were slightly lower and RMSECV
was higher for model 1 (probably due to different conditions during
spectral data acquisition)
Doramaxx Consulting
On-tree data were used in model 1; data for 2 obtained in the laboratory
The NIR prediction of fruit moisture and oil content in intact olives has
been previously reported.
León et al. (2003, 2004) obtained calibration models accurate enough
to predict oil content and moisture with r values of 0.94 and 0.93
Cayuela et al. (2009) obtained variable predictive ability based
on the sample presentation reference laboratory method.
The best results provided r values of 0.83 and 0.88 and RER values of
7.8 and 11.8 for oil content and moisture.
Using a different portable instrument, Cayuela and Pérez- Camino
(2010) obtained r values of 0.78 and 0.76 and RER values of 10.6
and 10.3 for oil content and moisture respectively.
Doramaxx Consulting
The different cultivars evaluated in this work showed differences in lipid
synthesis both in total amount of oil formed and the period of time,
as well as the evolution of fruit moisture during ripening (Figures 3
and 4).
The prediction values for oil content and moisture by cultivar and
sampling date were closely correlated with reference
Doramaxx Consulting
Figure 1. Change in moisture and oil content during the ripening period.
Each point represents the mean value
of 24 samples (Trial 1) and 16 samples
(Trial 2); error bars indicate the SE
Doramaxx Consulting
Figure 2: Avg. Spectra and CoV of Olive fruit samples
Trial 1 = Black; Trial 2 = Grey
Doramaxx Consulting
Tables 1 and 2
Doramaxx Consulting
Figure 3: Reference values v. Predicted values
Doramaxx Consulting
Figure 4
Doramaxx Consulting
Comparison of NIR and lab for oil and
moisture
Doramaxx Consulting
Rapid, reliable analysis can contribute to process and quality
improvements in numerous ways. For example,
Assessment of raw olive acceptability.
If the olives have been collected from the ground rather than fresh from
the tree, they may be of poor quality with high acidity and hence
lower value.
Measurement of water and oil content. These parameters determine
the price of the olives, with those having a greater oil content
commanding a higher price.
Process optimization. After extracting the oil, the remaining pulp or
by-product (called alperujo) should have only minimal oil content,
typically around 2% or less.
If the oil exceeds this level, a problem with the process is indicated.
Doramaxx Consulting
Methodology
Olive samples were milled to a paste and placed in a glass petri dish
before analysis.
Spectra were collected between 10000 and 4000 cm-1 at 16 cm-1
resolution, with an accumulation time of 30 seconds per sample.
The olive samples were also analyzed for oil and water content
following the customer’s established laboratory procedures.
Some of the measured spectra are shown in Figure 1.
Typically for NIR spectra, the absorption features are broad and
overlapped, although several prominent features can be assigned
either to water or to organic C–H modes in the oil.
Doramaxx Consulting
Figure 1: A number of olive samples
Doramaxx Consulting
CH = oil bands
Building the Software Model Using Principal Components
1. Define materials and acquire spectra of known references.
2. Optionally, configure algorithm parameters and spectral pre-
processing such as baseline correction
3. Calibrate the method.
The software automatically builds the models and determines the acceptance
thresholds.
4. Review the classification results (for example, see Figure 2).
Any issues with the data or performance of the method will be flagged by the
troubleshooting engine, allowing corrective action to be taken.
5. The validated method is then deployed as a workflow within the
dedicated Analyzer module, allowing routine use of the method.
Doramaxx Consulting
Quantitative Modeling of Oil and Water
The oil and water content of the olives are key parameters for quality
and both contribute to the NIR spectrum.
The complex nature of NIR spectra sometimes makes it impossible to
develop quantitative models based on the absorbance at a single
wavelength.
However, multivariate (Chemometric) methods such as partial least
squares regression (PLS) still function in the presence of
overlapping bands, and will allow models to be built.
The olive spectra and properties determined by chemical analysis were
loaded into software.
One third of the data were designated as a validation set to verify the
performance of the model.
The spectra were pre-processed with first-derivative baseline
correction.
Doramaxx Consulting
Table 1: Properties of the olives for water and oil
The calibration and validation results are summarized in Table 1 and
Figure 3.
The models use a modest number of latent variables and show good
linearity and precision over the range of available samples.
The standard errors of prediction (SEPs) were 1.5 % and 1.7 % for oil
and water, respectively.
Doramaxx Consulting
Experimental Results
Oil Water
Doramaxx Consulting
# PCs v. SEP
PLS Calibration
PLS Validation
Analysis Scheme
Doramaxx Consulting
What can be analyzed by NIR?
Doramaxx Consulting
Techniques used for each assay
Doramaxx Consulting
Results of Analyses
Olive Leaves
Soils
Olive fruit (intact)
Olive Paste
Doramaxx Consulting
Instrument Specs
Courtesy Unity Labs Doramaxx Consulting
Olive Oil Spectra
Doramaxx Consulting
Calibrations for Fat and Moisture
Fat
Moisture N = number of samples
RSQ = Correlation Coefficient (NIR vs. wet chem)
Min = minimum reference value
SECV = Cross Validation Error
Doramaxx Consulting
Laboratory Measurements
Doramaxx Consulting Courtesy Bruker Instruments
Spectra of EVOO, Soybean oil, EVOO adulterated
w/50% soy, pure “all natural” OO
Doramaxx Consulting
FT-NIR Analysis of Edible Oil Quality
FT-NIR spectroscopy can help to:
• Identify the incoming oil
• Assess the quality of the oil
• Evaluate the frying capabilities
Analysis:
• No sample preparation
• filling of 8mm disposable vials
• Temperature control at 50°C
• Measurement time: approx. 20 sec
Doramaxx Consulting
37
Property Data Set Prediction
Error
Name Unit n Min Max RMSEP
C16:0 Palmitic Acid % 608 4 16 0.50
C18:0 Stearic Acid % 673 1.7 92 0.98
C18:1 Oleic Acid % 800 0.1 85.2 0.65
C18:2 Linoleic Acid % 673 0.1 63.2 0.39
C18:3 Linolenic Acid % 415 0.1 9.2 0.14
Typical Performance for NIR Analysis to
Check the Quality of Fresh Edible Oils
Doramaxx Consulting
38
Property Data Set Prediction
Error
Name Unit n Min Max RMSEP
Gardner Color 766 0.1 4.1 0.35
TFA Trans Fatty Acids % 659 0.1 60.4 0.61
TFA low range % 265 0.1 2.9 0.11
IV Iodine Value 1056 0.2 195 0.93
FFA Acidity % 1193 0.1 7.6 0.09
Typical Performance for NIR Analysis to
Check the Quality of Fresh Edible Oils
Doramaxx Consulting
FT-NIR Analysis of oils & fats in
transflection
• Calibration development in progress
• Calibration from room temperature to
50oC
• Lower but still sufficient accuracies
• Probably less possible components or
parameters to be calibrated
Doramaxx Consulting
Hand-Held Models are Available microPHAZIR RX Handheld NIR Material Analyzer (a) and QualitySpec Trek (350-2500 nm)
Hand-held Spectrometer (b)
Doramaxx Consulting
a b
Courtesy Ahura [a] (Thermo) and ASDI [b]
Area NIR using “Drones” Whole fields or groves can be scanned
Doramaxx Consulting
LandSat Images showing
Productivity
Doramaxx Consulting
And then there’s dairy products
Doramaxx Consulting
QC of Cheese by NIR
Fat 7.5 - 35.0 +/- 0.3%
Dry matter 27.5 – 64. 5 +/- 0.5%
Doramaxx Consulting
Fat Dry Matter
On line analysis of Cheese Curd
Doramaxx Consulting
Transmission spectra
Doramaxx Consulting
Absorbance
Spectra
Derivative
Spectra
NIR Spectra of Cheese Curd
Doramaxx Consulting
Absorbance
Spectra
2nd Derivative
Spectra
Water Calibration
Doramaxx Consulting
Dry Matter and Fat in Cheese
Dry Matter Fat
Cheese Standards
Doramaxx Consulting
Typical Instrument Spec Sheet
Doramaxx Consulting
Conclusions
Because of its versatility (many types of sample
presentations), NIR is quite useful in food and
agriculture
With over 60 years of experience, there are many, many
references for food applications available, either in a
library or through instrument manufacturers
Instruments range from sophisticated laboratory to rugged
hand-held or instrument/vehicle mounted monitors
The speed of analysis allows for may more samples to be
(non-destructively) measured, assuring a good cross-
sample of a field or grove
Doramaxx Consulting