Università degli studi della Tuscia
Published September 21, 2020 | Version on line
Journal article Restricted

Pine nut species recognition using NIR spectroscopy and image analysis.

Authors/Creators

  • 1. Department for Innovation in Biological, Agro-food and Forest System, University of Tuscia, Via S. Camillo de Lellis Snc, 01100, Viterbo, Italy
  • 2. Department for Innovation in Biological, Agro-food and Forest System, University of Tuscia, Via S. Camillo de Lellis Snc, 01100, Viterbo, Italy Department of Natural Resources Management, Adigrat University, P. O. Box 50, Adigrat, Tigrai, Ethiopia
  • 3. United States Department of Agriculture, Agricultural Research Service, Western Regional Research Center, 800 Buchanan St, Albany, CA, 94710, United States
  • 4. Department of Agricultural and Forestry Sciences, University of Tuscia, Via S. Camillo de Lellis Snc, 01100, Viterbo, Italy

Description

Abstract

NIR spectroscopy and physical properties derived from image analysis were evaluated as potential features for the classification of seed kernels from two pine nut species (P. pinea L. and P. sibirica Du Tour) using Partial Least Squares Discriminant Analysis (PLS-DA). Model performances were evaluated in terms of specificity, sensitivity and accuracy. Data pre-treatments were essential for achieving excellent performances (accuracy rate > 95%) in all tests. The interval PLS-DA highlighted that the most important features for (1) the NIR method were the absorption bands at 1640–1658, 1720–1738 and 1880–1998 nm, while for (2) the image analysis were kernel eccentricity, kernel major axis length, kernel lightness (L*) and kernel perimeter. The results demonstrate potential of both techniques for discriminating the two pine nut species.

Notes

Highlights • Computer vision is feasible for pine nuts species recognition. • Image analysis ensures excellent results through shape and size measurements. • NIR spectroscopy shows classification performances similar to image analysis. • Classification accuracy is higher than 95% for each tested model. • The suggested approach provides the basis for a rapid online detection system.

Files

Restricted

The record is publicly accessible, but files are restricted to users with access.