Developing an alternative classification method for predicting ham composition using linear measurements from the cross-sectional ham surface

Abstract

Digital image analysis based on the ham cross-sectional face was used to measure two lean muscle and three subcutaneous fat locations from 248 bone-in hams. Linear measurements of the two selected fat locations were used to predict dual-energy X-ray (DXA) fat or lean percentages with prediction accuracies (R²) of 0.7 in a stepwise regression eq. A classification system was built based on the prediction equations, and the linear measurements aimed to classify extremes at the threshold of the 10th percentile of DXA fat percentage (> 32.0%) and lean percentage (< 60.2%). When using either DXA fat or lean percentage, lean ham prediction accuracy dropped by 18%, but fat ham prediction accuracy increased by 60% when the threshold was changed from the 10th percentile to the 30th percentile. This classification approach has the potential to be converted into a manual tool with several useful applications for commercial pork processors.

Keywords: Classification; Composition; DXA; Ham; Linear measurements.