Use of indirect multivariate calibration for quality control of agricultural products by near infrared spectroscopy

Abstract

In the near infrared (NIR) region, (λ=800–2500 nm) of the spectrum, most organic molecules have weak but reproducibly measurable absorption bands. This phenomenon has been increasingly exploited for the rapid, quantitative analysis of major constituents of agricultural products. As there is strong spectral overlap and an interaction of constituents in NIR, the calibration of instrumentation has to be performed indirectly with a training set of samples applying multivariate methods. A software package for IBM compatible personal computers includes programs to select samples to be used for calibration, to compute multiple linear regression equations, to cross-validate regression equations and to detect unusual samples. These methods are applied to rapidly determine the oil content of over a thousand rapeseed samples of the harvest of 1986. A considerable saving of time, labour, and costs can be achieved, the agreement with the standard laboratory method is very satisfactory.