Feasibility quantitative analysis of NIR spectroscopy coupled with Si-PLS to predict total acidity of <i>Seedless White</i> table grapes

Total acidity (TA) is a critical parameter for evaluating the quality of table grapes. Research on rapid detection techniques for table grapes contributes substantially to the comprehensive assessment of grape quality. This study employed visible-near-infrared (Vis-NIR) spectroscopy to rapidly and quantitatively determine TA in <italic>Seedless White</italic> grapes. Various spectral preprocessing techniques were employed on the spectral data within the 400 to 1100 nm wavelength range. The synergy interval partial least squares (Si-PLS) method was utilized to screen the optimal subintervals from the preprocessed spectral data correlating with the TA content in grapes. Spectral prediction models for total acidity were developed based on full-band spectrum data and optimal subintervals. The impact of various preprocessing methods on the accuracy of the TA prediction models was evaluated, and the performance of the full-band spectrum model was compared with that of the subinterval-based model. Through comparative analysis, the first derivative method combined with the Savitzky-Golay smoothing method emerged as the most effective preprocessing approach. Si-PLS was subsequently employed to select spectral intervals, and a prediction model based on these intervals was established. The optimal model showed a correlation coefficient (Rc) of 0.915 and a root mean square error (RMSEC) of 0.584 g/L for the calibration set, and a correlation coefficient (Rp) of 0.835 with root mean square error (RMSEP) of 0.788 g/L for the prediction set, yielding a residual predictive deviation (RPD) of 1.815. The results demonstrate that integrating NIR spectroscopy and Si-PLS facilitates the rapid and precise quantitative detection of TA in grapes. This study provides a reference for developing rapid detection devices.