www.qub.ac.uk/igfs - PowerPoint Presentation

Presentation on theme: "www.qub.ac.uk/igfs"— Presentation transcript

Slide1

www.qub.ac.uk/igfs

DEVELOPMENT OF A NOVEL CONTINUOUS STATISTICAL MODELLING TECHNIQUE FOR DETECTING THE ADULTERATION OF EXTRA VIRGIN OLIVE OIL WITH HAZELNUT OIL BY USING SPECTROSCOPIC DATA

Konstantia Georgouli1, Jesus Martinez Del Rincon2, Anastasios Koidis1

1Institute for Global Food Security, School of Biological Sciences, Queen’s University of Belfast, UK2Institute of Electronics, Communications and Information Technology, School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, UK

INTRODUCTION

Extra virgin olive oil (EVOO) is a premium vegetable oil characterised by great nutritional value and high price.Despite strict limits defining the purity of EVOO by International Olive Council (IOOC) and EU, it continues to attract various fraudulent and adulteration practices. Adulteration of EVOO with other vegetable oils is a certain problem that has not found yet solutions (European Commission 2013). Detection of adulterants at low levels (5-20%) is still difficult process (Zhang et al. 2011). Addition of hazelnut oil to extra virgin olive oil is one of the most concerning adulterations (Parker et al. 2014).

EXPERIMENTAL AND METHODOLOGY

REFERENCES:European Commission 2013, Workshop on olive oil authentication, European Commission, http://ec.europa.eu/agriculture/events/2013/olive-oil-workshop/newsletter_en.pdf.He, X. & Niyogi, P. 2004, "Locality preserving projections", Advances in Neural Information Processing Systems.Parker, T., Limer, E., Watson, A.D., Defernez, M., Williamson, D. & Kemsley, E.K. 2014, "60 MHz 1H NMR spectroscopy for the analysis of edible oils", TrAC Trends in Analytical Chemistry, vol. 57, no. 0, pp. 147-158.Zhang, X., Qi, X., Zou, M. & Liu, F. 2011, "Rapid Authentication of Olive Oil by Raman Spectroscopy Using Principal Component Analysis", Analytical Letters, vol. 44, no. 12, pp. 2209-2220.

ACKNOLEDGEMENTS: This research was funded by

MOTIVATION, METHODOLOGY AND RESULTS

AIM OF THE

STUDY: To develop a novel dimensionality reduction technique as a part of an integrated pattern recognition solution capable of identifying hazelnut oil (HO) adulterants in extra virgin olive oil at low percentages based on spectroscopic chemical fingerprints.

Creation of admixtures

(i)

Great Nutritional Value

High-priced food

Extra Virgin Olive oil adulteration

FTIR spectra acquisition

RAMAN spectra acquisition

Training dataset

Testing

dataset

1

.

Model the mixtures as data series

2. Mapping test samples on the model space

3.

Application

of a classifier

4. Validation of the model

Decision

Model

Exploratory

analysis

using

PCA, LDA

and

Kernel PCA

Data acquisition

Projection of

the produced

admixtures

on

the

space of

the

pure

oils

(Fig. 1)

Pretreatment

MOTIVATION

Figure

1.

Projection of the produced admixtures on the LDA space of the pure

oils

using FTIR data

METHODOLOGY

Continuous

Locality

Preserving

Projections (CLPP)

Based

on that conclusion, we developed

a

NOVEL statistical

technique modelling the produced admixtures

as data series instead of discrete

points

It extends

the linear dimensionality reduction technique Locality Preserving Projections, LPP (He,

Niyogi 2004). CLPP considers the mixture percentage as a continuous variable. Data is modelled as data series and the continuity preserved during the learning and dimensionality reduction.

Definition of CLPP

: Y={yk}(k=1..n), yk RD Z={mk}(k=1..n), mk Rd d<<DComplementary constraints control the similarity in the embedded space: neighbourhood graphs.

Design of our pattern recognition solution

Statistical analysis of the in house admixtures

Application of

CLPP

RESULTS

Figure

2.

Application of CLPP technique to RAMAN data

Continuous graphSimilarity graphThus, similarity and continuous neighbours are placed nearby in the embedded space without enforcing any artificial embedded geometry.Continuous neighbourhood : Similarity neighbourhood : The embedded space is spanned by the d eigenvectors with the smallest nonzero eigenvalues. They are obtained from the solution of the generalised eigenvalue problem:

Result

: a continuous reduced latent space where calibration data can be easily understood and analysed.

CONCLUSION

N

ovel dimensionality reduction approach, CLPP allows the preservation of the concentration grade information in the modelling of spectroscopic datasets

A

ddressing more

efficiently and accurately the subtle fraud of the adulteration of EVOO with HO based on spectroscopic datasets.

Conclusion

:

the admixtures have

continuous

nature.