Calibrations
The NIR instrument is not “calibrated” like a balance where the readings are merely adjusted up or down to a standard value.
The instrument has to be trained to recognise different products and constituents. This process of “training” is called the calibration procedure and herein lies the secret of success of this revolutionary technology.
For the training, a number of samples are analysed by traditional chemical analytical methods to determine the actual composition of the samples. Each of these samples is further placed in the NIR instrument and the reflectance values from the different wavelengths are obtained. With the aid of a microcomputer and powerful chemometric software the combination of analytical results and reflectance values are transformed to the calibration constants. This software is so powerful that great care must be taken that it does not merely present a statistical solution, but actually supplies a scientific solution that can be verified.
To develop any new calibration or even for maintaining existing calibrations, it is important to first physically source an ideal set of samples. For each product the sourced sample set must include samples that represent as much of the variation of the analytical and nutrient components that can be expected. This set should ideally also contain samples representing the natural variation that can occur. This includes the variation in cultivars, growing areas, growing conditions and growing seasons. Dersjant-Li and Peisker (2005) recently emphasised the large variation in nutritional composition between soya samples collected from different countries or even from different areas within the same country. Once a set of samples that covers most of the variation has been sourced, the majority of calibration software programs have a tool, which then aids in selecting a further sub sample set to prepare the calibration.
Furthermore, universal calibrations often supplied with the purchase of NIR instruments would rarely represent a true reflection of samples from local areas and usually need quite a bit of adjustment. This can be done by adding a number of carefully selected samples from a specific local product to the existing calibration data.