Near-infrared spectroscopy for detection of hailstorm damage on olive fruit

Authors: MOSCETTI, ROBERTI - University Of Tuscia; Haff, Ronald - Ron; MONARCA, DANILO - University Of Tuscia; CECCHINI, MASSIMO - University Of Tuscia; MASSANTINI, RICARDO - University Of Tuscia

Submitted to: Postharvest Biology and Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/9/2016
Publication Date: 7/5/2016
Citation: Moscetti, R., Haff, R.P., Monarca, D., Cecchini, M., Massantini, R. 2016. Near-infrared spectroscopy for detection of hailstorm damage on olive fruit. Postharvest Biology and Technology. 120:204-212.

Interpretive Summary: A rapid and economical computer program capable of identifying olives with hailstorm damage is important to the olive oil milling industry. Here, the feasibility of Near-Infrared spectroscopy for olive sorting (cv. Canino) into ‘Sound’ and ‘Unsound’ classes is reported. Three discriminant routines (Linear Discriminant Analysis, Quadratic Discriminant Analysis and k-Nearest Neighbor ) in combination with a genetic algorithm for feature selection (from 2 to 6 wavelengths) were tested to classify the olive fruits based on the correlation between their spectra and the presence of hailstorm damage. Spectral pretreatment and feature selection were optimized through an iterative routine developed in R statistical software. The performance of each discriminant model was defined based on false positive, false negative and total error rates. Selected models provided a very-good (< 5%) or good (<10%) total error rate for the predefined classes. The optimal features corresponded to R[1320 nm], R[~1460 nm], R[~1650 nm], R[~1920 nm], R[~2080 nm], R[~2200 nm] and R[~2220 nm]. The results indicate that single-point NIR spectroscopy could be feasible for hailstorm damage detection on olive fruit and on/in/at-line implementation on milling production lines for the achievement of both quality improvement and process standardization of the olive oil production.

Technical Abstract: A rapid, robust, unbiased and inexpensive discriminant method capable of classifying olive fruit (Olea europaea L.) on the basis of the presence of hailstorm damage is economically important to the olive oil milling industry. Thus, in the present study, the feasibility of Near-Infrared (NIR) spectroscopy for olive fruit sorting (cv. Canino) into two quality classes (i.e. ‘Sound’ and ‘Unsound’ classes, which correspond to not-damaged and damaged fruits, respectively) has been investigated. Three different discriminant routines (Linear Discriminant Analysis or LDA, Quadratic Discriminant Analysis or QDA and k-Nearest Neighbor or kNN) in combination with the genetic algorithm for feature selection (from 2 to 6 wavelengths) were tested to classify the olive fruits based on the correlation between their spectra and the presence of hailstorm damage. Spectral pretreatment and feature selection were optimized through an iterative routine developed in R statistical software. The performance of each discriminant model was defined based on a-error (false positive), ß-error (false negative) and total error rates. Selected models provided a very-good (< 5%) or good (<10%) total error rate for the predefined classes. The optimal features corresponded to R[1320 nm], R[~1460 nm], R[~1650 nm], R[~1920 nm], R[~2080 nm], R[~2200 nm] and R[~2220 nm]. Thanks to the results obtained, single-point NIR spectroscopy could be feasible for hailstorm damage detection on olive fruit and on/in/at-line implementation on milling production lines for the achievement of both quality improvement and process standardization of the olive oil production.