A diagnostic visible/near infrared tool for a fully automated olive ripeness evaluation in a view of a simplified optical system

A diagnostic visible/near infrared tool calibrated by means of image analysis,was proposed to evaluate the maturation degree of oil olives in order to replace traditional subjective methods in a view of future fully automated applications. Thirteen varieties of Olea europaea deriving from four regions of the south of Italy were analyzed. In order to objectify the ripening stage assessment, the RGB image was acquired. Spectroscopic analyses were performed using a benchtop FT-NIR and a portable vis/NIR instrument. The benchtop device was equipped with an optical fiber probe and the spectra were collected over the 800–2500 nm range, nominal resolution of 1.6 nm; the portable spectrophotometer cover the range of 400–1000 nm, nominal resolution 0.3 nm. The olive spectral data were modelled using Partial Least Squares - Discriminant Analysis (PLS-DA). The prediction capability reached by the global model (13 varieties were used) obtained from data acquired with both the devices results promising. The PLS-DA models calculated on the olives from Calabria, Sardinia and Abruzzo revealed high prediction capabilities, i.e. sensitivity, specificity and accuracy higher than 83%. The prediction capability of Apulia samples could be improved increasing the variability of the samples since for this region only 3 sampling times were considered. To compare the modelling performance between the benchtop FT-NIR and the portable vis/NIR device, a McNemar's test was performed resulting no significant difference between the PLS-DA global models. Finally, considering the good performance of the vis/NIR model, a variable selection using the interval PLS (iPLS) algorithm was applied. To reduce the complexity keeping the performance of the model built using the whole vis/NIR spectra (1647 variables), 12 bands (1.5 nm wide) were selected. The new model showed an improvement in terms of model stability and complexity (Sensitivity 86%; Specificity 87%; Accuracy 87%) than the two global models built with the whole vis/NIR and NIR range. The classification performance provided the groundwork for the development of (i) simplified systems for a direct olives ripening determination on the olive tree, and (ii) automated systems to be applied both in field and at the mill for olives sorting according to the ripening degree.