An improvement of multiplicative scattering elimination method of different spectroscopic analysis for assessing complex mixtures

The introduction of visible near-infrared (VIS-NIR) spectroscopy provides a powerful tool for enhancing the accuracy and efficiency of food internal quality analysis. However, scattering effects caused by variations in sample physical properties often interfere spectral signals, compromising the model performance in quantitative analyses of complex mixtures. Herein, this study adopted 16 spectral preprocessing methods, including eight common preprocessing methods applied individually and eight fused with self- developed spectral ratio (SR) technique. Partial least squares (PLS) and Random Forest (RF) algorithms were performed to correlate the quantitative evaluation of the target parameters. For meat samples, SR combined with standard normal variate (SR-SNV) preprocessing yielded optimal results. PLS models achieved test set R² of 0.992 for moisture, 0.970 for protein, and 0.994 for fat, with corresponding RMSE of 1.004 %, 0.581 %, and 1.108 %. In citrus analysis, SR-AUTO preprocessing produced the best PLS model for acidity (test set R²=0.739, RMSE=0.665 %), while SR-SNV preprocessing performed optimally for sugar content (R²=0.733, RMSE=0.582 %). This study establishes a robust framework for rapid, accurate quantification of key internal quality indicators in food products.