Color image analysis and machine learning for classification and flavonoid-equivalent prediction of Citri Reticulatae Pericarpium

Background Citri Reticulatae Pericarpium (CRP) authentication and flavonoid-related quality evaluation commonly rely on chromatographic or spectroscopic techniques, which are accurate but require specialized instrumentation. This study aims to develop a standardized colorimetric image-based machine-learning workflow for rapid CRP origin-aging classification and flavonoid-equivalent prediction. Results Representative flavonoid monomers were used to compare chromogenic responses in aluminum chloride and sodium nitrite-aluminum nitrate systems. CRP samples were analyzed using standardized image acquisition, color reference normalization, fixed ROI extraction, engineered RGB descriptors, and inverse sRGB gamma correction for regression modeling. For CRP classification, a 480-record model-development set and a 120-record independent held-out validation set were constructed at the image-group level. Through the two chromogenic systems, RF classifier achieved held-out accuracies of 0.9500, 0.9917, and 0.9500 for geographical-origin, aging-year, and combined origin-aging classification, respectively. For flavonoid-equivalent prediction, gamma correction improved the RF external validation R 2 from 0.9502 to 0.9789 and reduced the external RMSE from 5.4186 to 3.5231. Spike-recovery values were 104.27 % ± 8.94 %, 102.07 % ± 6.36 %, and 96.74 % ± 6.45 % at the 200, 400, and 800 μg/mL addition levels, respectively. Significance The proposed workflow integrates chromogenic reaction, standardized image acquisition, color feature extraction, gamma correction, and machine learning into a low-cost analytical strategy for CRP authentication. The results indicate that the method can serve as a useful tool for rapid and efficient quality evaluation for CRP.