Performance specifications for consistent Model Performance of an instrument using an Interactance Optical Design in prediction of internal fruit attributes

Handheld spectrophotometer units are used in the application of assessment of intact fruit on the tree for attributes such as Brix, dry matter, or pigment content. However variation in performance of individual units can occur.

The unit used in this study operates in the visible and Herschel region, utilizing an interactance optical design in which light rays illuminating an annulus on the sample, scattering internally through the tissue. Light rays that re-emerge normal to the collimating lens are collected for the spectrometer. This study considers alignment of lamp to detector and repeatability measures (of a white tile and of a sample fruit) as metrics to determine overall unit performance in terms of prediction of fruit attributes.

The standard deviation of repeated scans of a reference PFTE tile can be used to calculate a repeatability metric. High values of such a metric are associated with poor performance of the instrument in terms of the RMSECV of a PLS model of a fruit attribute such as Brix or DM. The normal variation in this metric between individual instruments was not correlated to instrument attribute prediction (RMSECV). This metric does not account for the ratio of specular to diffusely scattered light emerging from the fruit received by the detector. A repeatability metric based on mean spectrum divided by the standard deviation on the spectral range 729-975nm (as used in PLS models of Brix and DM) from repeated scans of a real sample, such as an apple, was a stronger predictor of the hardware’s ability to generate a reliable PLSR model. Reasons for this association will be presented.