Three-phase spectral mapping and regression method for non-destructive prediction of Citrus Physicochemical Index (CPI)

Traditional single-indicator approaches fail to comprehensively assess citrus quality, hindering rapid inspection. This study introduces a Citrus Physicochemical Index (CPI) that integrates four key indicators—sugar, pH, dry matter content, and water content—through optimal feature and weight mapping. Trained models were employed as feature extractors, with parameter-derived mapping values normalized twice to balance scale differences, followed by weighted summation to construct the CPI. Prediction analysis showed that SVR-RF-UVE-fresh achieved the best performance in fresh-state samples (R² = 0.969), while SVR-IVSO-dry provided stable predictions in the dry state (R² = 0.654), and PLS-GA-difference yielded effective results in the difference state (R² = 0.814). Results demonstrate that spectral states significantly affect prediction, with fresh spectra generally outperforming dry and difference states. This method overcomes the limitations of single-indicator evaluation and establishes a rapid, nondestructive, and comprehensive approach for citrus quality assessment, offering a scientific basis for industry-level quality inspection. • CPI integrates sugar, pH, water, dry matter; R² = 0.969 in fresh citrus quality. • Fresh model best (R²=0.969); diff spectra improve dry state (R²=0.814) • Key bands: 450–500 nm (stable), 650–700 nm (chlorophyll), 900–1000 nm (moisture). • SVR-RF hybrid cuts RMSEP to 0.042; adapts well to spectral state variability. • 400–1000 nm hyperspectral imaging enables fast, non-destructive citrus assessment.