Detection of attribute XXX in fruit YYY using NIRS

K.B. Walsh
NIRS has been used in commercial fruit grading since the 1990s, but the scientific literature base has taken some time to explore issues relevant to this application. An increasing number of publications are now appearing, but some of these studies fail to advance knowledge in the field. This presentation reviews the basics required for an acceptable study, and current 'gaps'. The reference method itself should be characterised, e.g., in terms of repeatability. The spatial distribution (within a fruit) of the attribute of interest, relative to the volume of fruit 'optically' sampled, and the distribution within the assessed population, should be described. Reports should fully characterise the instrumentation used in terms of lamp type, detector type, wavelength resolution and repeatability, optical geometry (lamp-sample-detector). A logical approach to chemometrics should be demonstrated, in terms of assessment of pre-treatment options, wavelength selection, and chemometric procedure. If Partial Least Squares Regression is used, careful attention to the groups used in cross validation is needed. Importantly, multiple populations of fruit should be used in assessment. Simple division of a single population into a calibration and a validation set is inadequate. Statistical comparison of the results of different methods is recommended. For published work, spectra and reference data should be made publicly available.
Walsh, K.B. (2016). Detection of attribute XXX in fruit YYY using NIRS. Acta Hortic. 1119, 141-146
DOI: 10.17660/ActaHortic.2016.1119.19
https://doi.org/10.17660/ActaHortic.2016.1119.19
fruit, processing, colour, texture, segmentation, count
English

Acta Horticulturae