Development of a hands-free vision system for objective pork loin colour evaluation under commercial processing...

Accurate pork colour grading is essential for quality classification and market specification but is commonly based on subjective visual assessment, which is prone to operator variability and inconsistency. This study developed and validated a context-aware, hands-free Pork Loin Vision System (PLVS) for automated pork loin colour evaluation under commercial processing conditions based on Canadian Pork Quality Standards (CPQS). The system integrates wearable image acquisition of the intact ventral surface of the pork loin with automated region-of-interest detection and colour analysis based on CIELAB (L*, a*, b*) coordinates and colour-difference modelling. Using ΔE-based calibration with in-image reference standards, the algorithm performs practical colour normalization to compensate for illumination and environmental variability. PLVS predictions were compared with trained grader classifications and Minolta CM-700d spectrophotometer measurements. Across colour classes (2-4), PLVS produced similar trends in mean CIELAB values with generally lower variability than subjective grading. Lightness (L*) decreased and redness (a*) increased with increasing colour class for all methods. Top-1 classification agreement with subjective grading was 74%, increasing to 98% when the top two predicted grades were considered, indicating that discrepancies were primarily limited to adjacent classes. Regression analyses showed stronger linear relationships between PLVS-derived colour values and class order compared with instrumental measurements. These findings demonstrate that region-based wearable computer vision can provide consistent and objective pork colour estimates and may serve as a practical decision-support tool for grading under commercial conditions.