Spectral Calibration for In-situ Measurement of Glycerol Fermentation Component Concentrations via Infrared Spectroscopy

For in-situ measurement of important component concentrations during a glycerol fermentation process via the near-infrared spectroscopy (NIR), this paper proposes a semi-supervised spectral calibration model building method based on the goose optimization algorithm (GO) for spectral data analysis to enhance in-situ measurement accuracy. To address the issue of very limited labeled samples as often encountered in practice, a semi-supervised approach based on combining the consistency regularization and pseudo-labelling strategy is presented, along with screening out reliable unlabeled spectra based on a wilcoxon signed-rank test for calibration model building. In addition, a comparative study on measuring the multiple component concentrations during a glycerol fermentation process is conducted in terms of using different modelling methods. Experimental results well demonstrate the measurement accuracy of the proposed spectral calibration method.