A VOTCA-based visible near-infrared spectroscopy modeling approach for addressing sample variability and production lines disparities in food quality analysis

Variations in sample characteristics across harvest periods, along with inconsistencies in spectrometer components, often compromise model transferability and robustness. To address this challenge, a novel model updating (MU) strategy called Variable Optimization Transfer Component Analysis (VOTCA) is proposed, aimed at constructing universal models with enhanced adaptability. In this study, the visible near-infrared (Vis/NIR) spectral of snow peach samples collected across seven harvest periods and two sorting lines were analyzed to predict soluble solids content (SSC). Two recognized MU methods including piecewise direct standardization (PDS) and spectral space transform (SST) were used to compare the results with the VOTCA. In harvest period transfer scenarios, VOTCA achieved better prediction accuracy (R_p = 0.852; RMSEP = 0.675; RPD = 1.910), while in cross-instrument transfer tasks, it demonstrated better performance (R_p = 0.906; RMSEP = 0.683; RPD = 2.365), outperforming PDS and SST. Moreover, VOTCA does not require additional calibration sets when predicting new batches, making it a flexible and efficient approach for real-time quality assessment in dynamic production environments.