Integrating computer vision descriptors with instrumental measurements to characterize color and texture evolution in whey–aguamiel cheese during refrigerated storage

INTRODUCTION: Monitoring the evolution of quality in dairy products often relies on instrumental measurements, whereas image-derived information is often analyzed separately, despite its potential to capture complementary visual changes. This study evaluated whether computer vision descriptors can be integrated with conventional instrumental measurements to characterize formulation-dependent evolution in a whey–aguamiel cheese system during refrigerated storage.

METHODS: Four cheese formulations differing in medium fermentation and Lacticaseibacillus casei supplementation were stored at 4 °C and evaluated over 28 days. Instrumental color (CIE L*a*b*) and texture profile analysis (TPA) parameters were measured, while top-view images were processed to extract HSV color metrics and texture descriptors derived from gray-level co-occurrence matrices (GLCM) and local binary patterns (LBP). Instrumental and image-derived datasets were integrated using treatment × time as the analytical unit. Correlation analysis and principal component analysis (PCA) were applied to identify cross-modal relationships and to describe product evolution within a reduced multivariate space.

RESULTS: Instrumental data showed formulation-dependent changes in lightness and hardness, with non-fermented treatments presenting higher hardness values. Image-derived features varied across treatments and storage time: Hue reflected chromatic shifts, whereas GLCM contrast and LBP entropy captured surface heterogeneity at different texture scales. Correlation analysis revealed structured links between visual descriptors and instrumental attributes. PCA summarized 49.5% of the total variability in the first two components, separating samples by formulation and time. Non-fermented treatments occupied more dispersed regions in the PCA, while fermented treatments showed more constrained trajectories.

CONCLUSIONS: Image-derived color and texture descriptors were coherently integrated with instrumental measurements in a shared multivariate space. This integration showed that the evolution of whey–aguamiel cheese during refrigerated storage could be interpreted as formulation-dependent visual–structural trajectories. The results support computer vision as a complementary, non-destructive source of information for multivariate characterization of dairy product evolution.