Band Target Entropy Minimization and Target Partial Least Squares for Single Target Multivariate Curve Resolution and Calibration
share
Band target entropy minimization (BTEM) and target partial least squares calibration (T-PLS) were employed to resolve estimates for single pure component spectra and to calibrate those components concentrations in a true one-at-a-time fashion in unknown mixtures. It was found that BTEM could be used to provide spectral estimates that, when applied as the target spectra in T-PLS afforded similar errors of prediction as those from a standard multivariate curve resolution method, multivariate curve resolution alternating least squares (MCR-ALS). The calibration of BTEM resolved spectral estimates by T-PLS rather than the typical classical least squares calibration allows for the semiquantitation of individual well-represented components when other components of an unknown mixture do not have sufficient independent variation for spectral resolution. Two experimental scenarios were investigated to demonstrate situations where the combination of BTEM and T-PLS could be applied to model the pure component spectra and concentration profiles and MCR-ALS could not. The combination of both techniques are shown to resolve and calibrate a chemical impurity, melamine, by reflectance measurements of serially diluted samples of reference milk powder and of dimethyl methylphosphonate from a mid-infrared hyperspectral image of a gas stack.
READ FULL TEXT