Minimisation of instrumental noise in the acquisitionof FT-NIR spectra of bread wheat using experimental designand signal processing techniques

Spectral resolution (R) and number of repeatedscans (S) have a significant effect on the S/N ratio of Fouriertransform-near infrared (FT-NIR) spectra, but the optimalvalues of these two parameters have to be determinedempirically for a specific problem, considering separatelyboth the nature of the analysed matrix and the specificinstrumental setup. To achieve this aim, the instrumentalnoise of replicated FT-NIR spectra of wheat samples wasmodelled as a function of R and S by means of the Doehlertdesign. The noise amounts in correspondence to differentexperimental conditions were estimated by analysing thevariance signals derived from replicate measurements withtwo different signal processing tools, Savitzky–Golay (SG)filtering and fast wavelet transform (FWT), in order toseparate the “pure” instrumental noise from other variabilitysources, which are essentially connected to sample inhomogeneity.Results confirmed that R and S values leading tominimum instrumental noise can vary considerably dependingon the type of analysed food matrix and on the differentinstrumental setups, and helped in the selection of theoptimal measuring conditions for the subsequent acquisitionof a wide spectral dataset.