Modeling of Backfat on Untrimmed Pork Loins

Abstract

The current practice of manually trimming backfat from fresh pork loins is both labor intensive and inaccurate. If the fat thickness over the entire loin could be accurately predicted by measuring a few specific points, this trimming process could be mechanically automated. A model was developed for predicting the backfat profile of pork loins based on fat thicknesses at several discrete locations. Machine vision was used to create images of 52 loins, cut into 2.5 cm (1 in.) thick slices. Fat thickness was determined at four locations (T1-T4) on each slice, using image processing. Thickness T1 was located at the belly edge and T4 was located at the dorsal midline in relation to the carcass. With data from 32 loins, SAS(TM) was used to model each of these four fat locations along the length of the loin. The resulting model uses four known fat thicknesses at each location (T1-T4), for a total of 16 measured points on each loin. When the model was validated on 20 loins, the average coefficients of determination between actual and predicted thicknesses were 0.62, 0.66, 0.59, and 0.59 at locations T1, T2, T3, and T4, respectively. More importantly, average predicted cutting positions were within 0.20 cm (0.08 in.) of the desired position with SD < 0.83 cm (0.33 in.), resulting in minimal losses. This indicates that the similarity among loin fat thickness patterns may be great enough to permit the use of a model for automating the trimming process.