Evaluation of the computer vision system (CVS) to predict primal composition of mature cows

Yield prediction using rib-eye or cold carcass cameras (CCC) is not achievable on mature animals. This study
was undertaken to determine the potential of the computer vision systems (CVS) for the whole-side or hot carcass camera (HCC) to predict primal composition (fat, lean, and bone) of mature cows. A total of 111 mature cows were slaughtered at the AAFC-Lacombe Research and Development Centre federally inspected abattoir. Immediately after slaughter, pictures of each carcass side were taken using a HCC
(VBS 2000, e+v® Technology GmbH, Germany). Following 72 h of chilling at 2°C, carcass sides were weighed and knife ribbed between 12th-13th ribs. After 20 min of atmospheric exposure, CCC (VGB 2000 e+v® Technology GmbH, Germany) pictures at the grade site from left and right rib-eyes were taken. All images were analyzed using camera software to yield 166 and 99 variables from the HCC and CCC, respectively. Left carcass sides were fabricated into primal cuts, with carcass breakpoints identified following the Institutional Meat Purchase Specifications for Fresh Beef Products: Chuck (#113), rib (#103), loin (#172A) and round (#158A). Then, all primals were fully dissected into fat, lean and bone, and weighed by qualified personnel. Partial least squares regression procedure was used to predict primal composition using
the CVS data as independent variables. The accuracies were assessed by the coefficient of determination (R2) and root mean square error of cross-validation (RMSE).
Overall, proportion of variation in primal composition accounted for by HCC prediction equations was very high
across the primal cuts studied, Most of the lean and fat predictions for the primal cuts showed high coefficients
of determination, being R2 values for fat (R2~0.85 on average) slightly higher than those for lean (R2~0.75 on average). Specifically, HCC predictions showed higher R2 values for the fat composition of rib (0.87) and round (0.85) than those observed for CCC with limited improvement by using data from both grading cameras (HCC+CCC). In the case of lean, HCC improved the R2 values for chuck (R2=0.85), loin (R2=0.82), and round (R2=0.90) compared to the CCC, but superior prediction was obtained for the rib using the CCC (R2=0.69) and the combined CCC+HCC (R2=0.79). Overall, observed R2 values for predicting bone in the primals were lower (R2=0.36-0.79) using the HCC than those for lean and fat. The CCC camera alone had limited ability to predict bone (R2=0.09-0.38) but when combined with the HCC camera predictions for bone in the chuck and the loin were slightly improved (R2=0.71 and 0.76, respectively). On average, the explained variability with HCC was 3.32%, 24.4% or 61.8% higher than that observed with CCC for fat, lean, or bone, respectively.
Early literature has reported the suitability of CCC to predict total lean or saleable carcass yields. However,
the current preliminary results suggest that individual primal cut composition of mature cows can be accurately predicted by CVS using the HCC alone. This is an important finding for slaughter systems, such as those used for mature cattle in Canada, that do not routinely knife rib carcasses which negates the use of CCC.