Optics-based method for enhancing spectroscopic calibration model development

Spectrometric calibration is a challenge for every application because a large number and variability of reference samples are required to develop an accurate and robust model. This study investigates the potential of improving spectrometer calibration by using physically simulated spectra to enhance the calibration data set and perform out-of-range predictions. For this purpose, 56 liquid samples covering a wide range of absorption and scattering properties are measured with a double integrating sphere setup and visible spectrometers. A Kubelka-Munk optical model is used to simulate the samples reflectance and transmittance spectra. A simple method of calibration enhancement is introduced, and its benefits are evaluated with a partial least squares regression model, in comparison to a classical calibration approach. Two cases are studied, one with wide variability among reference samples and one with a limited range. The proposed method shows promising results, especially in the second, hardest case. A further study is done to evaluate the impact of simulated data quantity on the models' results. It was found that the performances reach a maximum when the ratio between measured and simulated spectra is 1:2.