Modeling effect of visible-near infrared spectrum on mutton glucose content based on SNV and MSC combined with genetic algorithm

To improve the stability and prediction ability of the visible-near infrared spectral model for mutton nutrients, taking glucose (GLU) as an example, the characteristic wavelength was extracted by genetic algorithm (GA) and a prediction model was established. Two preprocessing methods, standard normal transformation (SNV) and multivariate scattering correction (MSC), were used to directly model the partial least squares regression and the results were compared. Genetic partial least squares model under SNV (GA-SNV-PLS) was better than the direct partial least squares model under SNV (FS-SNV-PLS). After cross-validation, the root mean square error (RMSE) of the model was 0.122, determinant coefficient R² was 0.930, and relative analysis error (RPD) was 2.295. Compared with the full spectrum, the R² and RPD for MSC and genetic partial least square model under MSC increased by 95.80%, 50.21%, 85.05%; 62.65%, 37.08%, and 52.54%, respectively.