Internal Quality Assessment of Mandarin Fruit by vis/NIR Spectroscopy

Abstract

Different measurement modes (reflectance, interactance and transmission) and spectral windows (500-1100 nm) were compared for their ability to predict non-destructively the harvest soluble solids content (SSC) and titratable acidity (TA) of Satsuma Mandarin (Citrus reticulata cv. "Miyagawa"). A direct transmission measurement mode (source and sensor opposite sides of fruit) coupled with a single spectral window from 700 to 930 nm delivered the most accurate SSC predictions (R 2 = 0.93, RMSEP = 0.32%). Accurate TA prediction was not possible with any measurement mode and the predictive ability that was achieved (best results: R 2~0 .65, RMSEP~0.15%) was only afforded indirectly through a correlation with the skin chlorophyll changes that occur as the fruit mature.

Assessment of internal quality attributes of mandarin fruit. 1. NIR calibration model development

“…As such, detection using NIRS is unlikely and we agree with the assessment of McGlone et al (2003) that previous reports of calibration on this attribute are likely to represent secondary correlations on attributes related to fruit maturity.…”

“…Using the WINISI stepwise MLR, models generally defaulted to 4-9 wavelengths, including wavelengths close to 900, 870, and 860 nm for TSS and 907, 890, and 780 nm for DM. McGlone et al (2003) noted that 'vast numbers of different spectral windows can be created over the wavelength range and all possible options could not be investigated in any reasonable time'. These authors limited their investigation to 7 spectral windows chosen on the basis of 'prior experience and intuition', concluding that a wavelength range of 750-1100 nm was optimal for interactance spectra.…”

“…Similarly, Miyamoto et al (1998) used NIR transmittance spectra of intact Satsuma mandarins to achieve prediction of TA in separate populations (origin of each population not stated), with R v 2 of 0.83, bias of 0.02% TA, and RMSEP of 0.15% TA. However, McGlone et al (2003) concluded that robust TA prediction was not possible in Satsuma mandarin, although R v 2 of up to 0.65 and RMSEP of 0.15% TA could be achieved within a given population through a correlation with skin chlorophyll (fruit maturity).…”

“…The R c 2 (calculated in this instance as 1 − (RMSECV/SD) 2 ) for these values was 0.67. McGlone et al (2003) explored the use of several optical configurations and wavelength windows for prediction of TSS in Satsuma mandarin, reporting best results (R v 2 of 0.93 and RMSEP of 0.32% TSS) for a transmittance methodology operating between 700 and 930 nm. Results for an interactance geometry (R v 2 of 0.85 and RMSEP of 0.47% TSS) were superior to that for a reflectance geometry (R v 2 of 0.75 and RMSEP of 0.63% TSS).…”

Assessment of internal quality attributes of mandarin fruit. 2. NIR calibration model robustness

“…Unfortunately, most reports on the application of NIRS to fruit sorting describe the use of a single harvest population, divided into a calibration set and a validation set. For example, McGlone et al (2003) worked with 20 populations of mandarins (from 3 orchards over 7 weeks), with a calibration set assembled from 75% of each set and used to predict a population consisting of the remaining samples from each of the sets. This procedure allowed an estimate of prediction error; however, it did not involve testing the calibration against fruit from independent populations (e.g.…”