Near infrared spectroscopy is suitable for the classification of hazelnuts according to Protected Designation of Origin

• Published in: Journal of the science of food and agriculture Vol. 95; no. 13; pp. 2619 - 2625
• Main Authors: Moscetti, Roberto, Radicetti, Emanuele, Monarca, Danilo, Cecchini, Massimo, Massantini, Riccardo
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Ltd 01.10.2015; John Wiley and Sons, Limited
• Subjects: Agriculture; Algorithms; Calibration; chemometrics; Classification; Cluster Analysis; Corylus; Corylus avellana L; cultivars; Dietary Fats - analysis; Discriminant Analysis; Food Labeling; Hazelnuts; Humans; Infrared spectroscopy; Italy; Least-Squares Analysis; Models, Theoretical; non-destructive techniques; Nuts; Nuts - chemistry; Nuts - classification; Origins; Pretreatment; Routines; Species Specificity; Spectroscopy, Near-Infrared - methods; Spectrum analysis; Support Vector Machine; Support vector machines; Italy; non-destructive techniques; cultivars; chemometrics; discriminant analysis; Corylus avellana L
• ISSN: 0022-5142; 1097-0010
• DOI: 10.1002/jsfa.6992

BACKGROUND This study investigates the possibility of using near infrared spectroscopy for the authentication of the ‘Nocciola Romana’ hazelnut (Corylus avellana L. cvs Tonda Gentile Romana and Nocchione) as a Protected Designation of Origin (PDO) hazelnut from central Italy. Algorithms for the selection of the optimal pretreatments were tested in combination with the following discriminant routines: k‐nearest neighbour, soft independent modelling of class analogy, partial least squares discriminant analysis and support vector machine discriminant analysis. RESULTS The best results were obtained using a support vector machine discriminant analysis routine. Thus, classification performance rates with specificities, sensitivities and accuracies as high as 96.0%, 95.0% and 95.5%, respectively, were achieved. Various pretreatments, such as standard normal variate, mean centring and a Savitzky–Golay filter with seven smoothing points, were used. The optimal wavelengths for classification were mainly correlated with lipids, although some contribution from minor constituents, such as proteins and carbohydrates, was also observed. CONCLUSION Near infrared spectroscopy could classify hazelnut according to the PDO ‘Nocciola Romana’ designation. Thus, the experimentation lays the foundations for a rapid, online, authentication system for hazelnut. However, model robustness should be improved taking into account agro‐pedo‐climatic growing conditions. © 2014 Society of Chemical Industry