Nonlinear least squares with local polynomial interpolation for quantitative analysis of IR spectra

When using spectroscopic instrumentation for quantitative analysis of mixture, spectral intensity non-linearity and peak shift make it challenging for building calibration model. In this study, we investigated the performance of a nonlinear model, namely nonlinear least squares with local polynomial interpolation (NLSLPI). In NLSLPI, the parameters to be optimized are the concentrations of the components. Levenberg-Marquardt (L-M) method is used to solve the nonlinear-least-squares optimization problem and local polynomial interpolation is used to generate the nonlinear function for each component. We tested the robustness of NLSLPI on a computer-simulation dataset. We also compared NLSLPI, in terms of RMSEP, to partial least squares (PLS), classical least squares (CLS) and piecewise classical least squares (PCLS) on a real-world dataset. Experimental results demonstrate the effectiveness of the proposed method.