Construction of an online reduced-spectrum NIR calibration model from full-spectrum data

Near infrared (NIR) spectroscopy offers rapid and nondestructive estimation to a wide range of industries, but its acceptance has been slowed by the high costs of long-term use of full-spectrum instrumentation. From examining the terms produced in multivariate calibration of this full-spectrum data, it is possible to identify influential wavelengths, using either the regression vector b or a series of estimation prognostic vectors c, which is proposed in this paper. Once these wavelengths have been identified, the full-spectrum probe can be replaced with a series of monochromators, which is more commercially viable. In this paper, online NIR absorbance data from a pilot scale food extruder is used to estimate downstream product quality attributes (PQAs) via a full-spectrum calibration model followed by a reduced-spectrum model.