NIRS-Based Fresh Grape Ripeness Prediction with SPA-LASSO Spectral Feature Selection

Abstract

This study proposes a rapid, non-destructive maturity evaluation model was developed using Near-Infrared Spectroscopy (NIRS) to monitor dynamic quality parameters during grape ripening and identify the optimal harvest window. The study first characterized physicochemical changes in Cabernet Sauvignon grapes across twelve developmental stages to establish a foundation for predictive modeling. Feature wavelengths were extracted from five preprocessed spectral datasets using the Successive Projection Algorithm coupled with Least Absolute Shrinkage and Selection Operator (SPA-LASSO), followed by the construction of Partial Least Squares Regression (PLSR) models to predict Soluble Solids Content (SSC) and Total Acidity (TA).Key trends observed include a progressive increase in SSC and decline in TA from fruit enlargement to ripening, followed by a marginal SSC reduction and TA rise during late maturity. The SG+SPA-LASSO+PLSR model demonstrated superior performance, achieving coefficients of determination (R²) of 0.982 (calibration, R2C) and 0.983 (prediction, R2P) for SSC, with root mean square errors (RMSE) of 1.010 (RMSEC) and 0.978 (RMSEP). For TA, the model yielded R2C = 0.954, R2P = 0.944, RMSEC = 2.347, and RMSEP = 2.618. The SPA-LASSO method significantly improved feature selection efficiency, enhancing model generalizability for spectroscopic-based, non-destructive grape maturity assessment.