3D point cloud based optical tracking of dynamic quality degradation during drying of fruits and vegetables

Quality degradation during drying fruits and vegetables is mainly due to challenges in adaptation of advanced quality monitoring and control during processing. This research explores non-invasive and invasive measurements to analyze the quality of purple carrots, golden kiwifruit, blueberry and raspberry during drying. Quantitative changes in moisture content (MC), total anthocyanin (TA), ß-Carotene (BC), lutein, vitamin-C (VC), total phenolic contents (TPC) and total flavonoids (TF) along with colorimetric and physical changes were evaluated and compared with 3D point cloud based digital data. First-order kinetic model provided better fits (R BC 2 = 0.96, R VC 2 = 0.99, and R MC 2 = 0.96), confirming first-order degradation behavior during drying. The integration of kinetic modeling (zero- and first-order) with machine learning enabled accurate prediction of drying-induced quality changes in selected products. Hybrid gradient boost regressor (Hybrid-GBR) achieved the best results for MC across all products, with Rp 2 = 0.988 (RMSEP ≈ 0.037) for purple carrot, Rp 2 = 0.963 (RMSEP = 0.068) for raspberry, and Rp 2 = 0.980 (RMSEP = 0.041) for blueberry. For secondary metabolites (TA and TPC), hybrid Gaussian process regressor (Hybrid-GPR) and Hybrid-GBR models consistently outperformed conventional methods, resulting in test Rp 2 = 0.901, RMSEP = 50.4 (TA, raspberry) and Rp 2 = 0.867, RMSEP = 0.067 (BC, blueberry). For vitamin C in golden kiwifruit, Hybrid-PLSR performed best (Rp 2 = 0.905, RMSEP = 67.9). These findings show that low-cost 3D point cloud imaging can be integrated with spectral imaging with broader dataset for more accurate, real-time quality monitoring for dynamic optimization of drying process. • 3D point-cloud imaging applied for non-invasive quality monitoring during drying • First-order kinetics accurately described degradation of nutrients and moisture • Hybrid kinetic–machine learning models improved prediction of quality attributes • Hybrid-GBR and Hybrid-GPR achieved highest prediction accuracy across products