Robustness of Models Based on NIR Spectra for Sugar Content Prediction in Apples

Abstract

The sugar content of Golden Delicious apples is predicted using near infrared (NIR) spectrometry. The study focuses on the metrological characteristics of the sugar content measurement and external parameters involved in the lack of robustness of the NIR-based model. The external parameters were fruit temperature, spectrometer temperature and ambient light. The first two factors influenced the prediction accuracy: (i) a fruit temperature variation altered the prediction, the relationship seems to be described by a non-linear model within the considered temperature range, (ii) a variation of the spectrometer temperature also altered the prediction, the relationship is described by a linear function for a temperature between 4 and 30°C. Ambient light did not show to have any influence on the NIR-based model. The analysis of the metrological parameters showed a satisfactory repeatibility in sugar prediction with a low error, 0.073°Brix. The model reproducibility was good regarding bias-corrected standard error of prediction ( SEPc) without significant differences between experiments, on the other hand a bias remained even if the previous parameters were maintained constant. These results will be taken into account in future measurements, in order to improve the robustness of the NIR-based model developed for apples.

Application of LS-SVM to non-linear phenomena in NIR spectroscopy: development of a robust and portable sensor for acidity prediction in grapes

“…The 830-920nm spectral window, corresponding to the sugar absorbance peaks [22], was removed to avoid the concomitant influence of the major chemical constituent in grapes. For portable application test, only a few wavelengths were selected.…”

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

“…Both reports concluded that if the calibration model was developed with sample temperatures covering the range of future sample temperatures, then prediction accuracy was high. Sanchez et al (2003) also noted that the influence of spectrometer and fruit (apple) temperature was mainly on bias, not RMSEP(C). However, the effect of spectrometer temperature on bias was more than twice that of fruit temperature.…”

Non‐destructive measurement of vitamin C, total polyphenol and sugar content in apples using near‐infrared spectroscopy

“…Recent research on the use of near‐infrared spectroscopy (NIRS) on apples has focused on determining the soluble solids content, firmness and acidity . However, these are not the only quality attributes of relevance in apples.…”