Abstract
Journal of Near Infrared Spectroscopy
Volume 16 Issue 3, Pages 311–315 (2008)
doi: 10.1255/jnirs.793
Removing the block effects in calibration by means of dynamic orthogonal projection. Application to the year effect correction for wheat protein prediction
Jean-Michel Roger,a,* Fabien Chauchardb and Phil
Williamsc
aInformation and Technologies for Agro-processes–Cemagref, BP 5095, 34033 Montpellier Cedex 1,
France
bOndalys, SARL, Parc Minea–Cemagref, BP 5095, 34033 Montpellier Cedex 1, France
cPDK Projects, Inc., 5072 Vista View
Crescent, Nanaimo, British Columbia, Canada V9V 1L6
NIR spectroscopic models lack robustness towards the infuence of external parameters. Orthogonalisation techniques have demonstrated their ability to improve model robustness with regard to external parameter effects. The dynamic orthogonal projection (DOP) method is applied to the correction of the growing season effect on wheat protein content prediction using NIR spectroscopy. A few reference samples are measured at the beginning of each harvest. For each one, DOP generates an“ideal spectrum” by applying a reference-centred linear kernel on the calibration database. The parasitic sub-space, spanned by the differences between ideal and measured spectra, is then removed from the calibration space, by means of an orthogonal projection. The application database covered eight years of wheat NIR spectra related to protein content. The first four years were used for calibration and the remaining years for testing. The corrections were calculated using the first scanned samples of every year, for each variety [between five and ten samples (four samples per year)]. Results were compared with standard partial least squares calibration models.
Keywords: orthogonal correction, robust calibration, PLS, external parameter influence correction
Full-text article (429 kB) (subscribers only)
Buy article on-line for £113.75 (get immediate access)
Permalink: http://dx.doi.org/10.1255/jnirs.793
QR Code (what is this?):



