Non-destructive, portable, handheld spectroscopic devices for screening purposes

Transcription

1 Non-destructive, portable, handheld spectroscopic devices for screening purposes RIKILT Authenticity of nutrients Yannick Weesepoel SPICED Workshop on authenticity 8 th July 2015

2 Current trends in food authentication Chemist, not a Statistician! Link Simple, fast, on-site equipment for suspect sample discrimination ISO methods, definite analysis Results to be used in courts

3 Portable near infrared spectroscopy: Within seconds a tentative answer to an authenticity problem MicroNIRs: USB powered 60 grams (45 x 42 mm) Spectral range of nm Applications Harbours NVWA/FDA inspectors Supermarkets, retailers, etc.

4 Mode of action of near infrared spectroscopy IR spectrum Multivariate analysis toolbox Answer Time required: 5 s Costs: Nihil Confidence: %

5 Factor 2 Targeted approach vs multivariate analysis Targeted approach: single markers Protein content of milk Moisture/protein content of meat Multiple markers, untargeted analysis, chemometrics, multivariate analysis One sample! Factor 1 5

6 Examples of MicroNIR application development at RIKILT 1. Fraud detection with chicken meat non destructive 2. Other meat applications with support vector machines 3. Baselining of edible oils

7 Poultry meat production and import (EU28) Country Production 2012 (tonnes) France 1,925,336 United Kingdom 1,632,779 Germany 1,604,479 Poland 1,499,870 Italy 1,276,800 Year Import from non-eu , , (Jan-Aug) 558,590 (similar to 2013) COMEXT statistics Spain 1,103,595 Netherlands 1,041,014 (non-eu) Hungary 444,959 Belgium 366,991 Portugal 313, EU 28 total 13,222,369 FAO statistics Port of Rotterdam

8 Predicted value Absorbance Pilot experiment Create your own fraud samples! 1. Samples Fresh Thawed Freezing (-24oC) for 24 h MircoNIR measurements De-frosting for 24 h Samples 1.3 Fresh Gelatin injected Brine injected 24 h. Injection of solution dispersion of Wavelength (nm) solution Chemometric analysis 3. Reference methods for: Moisture content (NEN-ISO 1442:1997) Protein content (NEN-ISO 937:1992) Reference value

9 Raw NIR spectra

10 Parallel Absorbance Non-destructive measurements are a direct cause for random data Spot-to-spot variation Background Fibre direction Wavelength (nm) Counter parallel Parallel Counter parallel Wavelength (nm)

11 Support vector machine classification New generation self learning algorithms: 50% of set for building model 50% of set for repeated external validation Actual Predict. Fresh Injected Thawed Fresh Injected Thawed

12 Support vector machines (SVM)? Chris Pederson, JDSU Corp.

13 Self learning quantitative models Support vector machine regression Ext. val. set Moisture (%) Protein (%) Moisture/Protein (-) Average difference of actual and predicted values Average

14 Poultry meat authentication 2.0 Measuring through packages for routine retail or warehouse check: validation of first set of laboratory experiments in collaboration with retailers

15 Program the models in a user friendly interface for inspectors

16 First pilot with fish fresh/thawed Develop and confirmed results with HADH enzymatic essay

17 Benchmarking of edible oils for fast authentication?

18 Measuring through bottles?

19 Other handheld, portable spectroscopic techniques (1) Handheld-Raman spectroscopy Comparable to Near Infrared: both fingerprinting techniques for composition information Complementary to NIR (on molecular level) Raman has less interference with moisture in watery products Raman is especially suitable for minerals and pharmaceutical products Raman measurements are more difficult to standardize

20 Other handheld, portable spectroscopic techniques (2) Handheld-hyperspectral camera RIKILT in 2015 Visual of NIR Hot-spots, additions, mixtures RIKILT: Milk powder (broad anomaly testing) Saffron (fake/genuine/mixtures) Beef steak (organic/conventional)

21 Thank you for your attention! Questions?