Multivariate Quantitative Prediction of Soluble Solids Content, Moisture Content, and Fruit Firmness in ‘Dinosaur Egg’ Apricot Plum via Near-Infrared Spectroscopy with Cross-Parameter Feature Fusion and SHapley Additive exPlanations-Based Optimization

To meet market demand for fresh ‘Dinosaur Egg’ Apricot plum and realize effective quality classification, this study developed a non-destructive quality evaluation method using near-infrared spectroscopy (NIRS) with cross-parameter feature fusion. Spectral data were preprocessed, and key bands were screened via Competitive Adaptive Reweighted Sampling (CARS) and Shuffled Frog Leaping Algorithm (SFLA). Partial Least Squares Regression (PLSR) models for soluble solids content (SSC), moisture content (MC), and fruit firmness (FF) were established. Chemical index features were fused with FF-related preliminary features, and SHapley Additive exPlanations (SHAP) optimized feature contribution. Final models showed high performance: SSC (Rc2 = 0.9354, Rp2 = 0.9302, RMSE = 0.5212° Brix), MC (Rc2 = 0.9367, Rp2 = 0.9314, RMSE = 5.037 × 10−5), and FF (Rc2 = 0.8151, Rp2 = 0.7986, RMSE = 1.2710 N). This strategy improved the multi-quality detection accuracy, especially for FF, and provides technical support for intelligent fruit grading.