IMAGE ANALYSIS OF BEEF Table 3. Prediction equations for estimating retail product yield and longissimus area
aAbbreviations are defined in Table 2.
3.4 3.0 2.8
5.0 4.6 4.5
Development Equation Retail product yield, %
Predicted = 22.5 + (.79 Predicted = 16.6 + (.76 Predicted = 34.7 + (.67
PERLEANa) PERLEAN) + (.18 BLUE) PERLEAN) + (.015 NUMHOLES) +
( .039 HCW) Predicted = 47.0 + (.65
NUMHOLES) + ( .67
PERLEAN) + (.72 BLUE) + (.014 DENSITY)
Predicted = 41.4 + (.70 PERLEAN) + (.67 BLUE) + (.022 NUMHOLES) + ( .61 DENSITY) + ( .00092 HOLEAREA)
Longissimus area, cm2
Predicted = 6.2 + (.0018 LEAN) Predicted = 3.6 + (.0016 LEAN) + (.018 NUMHOLES) Predicted = 1.4 + (.0018 LEAN) + (.017 NUMHOLES) +
( .22 PERLEAN) Predicted = 159.0 + (.0029 LEAN) + (.019 NUMHOLES) +
( 2.7 PERLEAN) + ( .0020 FAT) Predicted = 159.4 + (.0031 LEAN) + (.020 NUMHOLES) +
( 2.7 PERLEAN) + ( .0018 FAT) + ( .052 HCW)
.86 .89 .87
.84 .85 .88
Retail Product Weight. We (Shackelford et al., 1995) have shown that RPWT can be predicted more precisely by multiplying predicted RPYD times HCW rather than by predicting RPWT directly. Thus, we predicted RPWT by multiplying the results of Equa- tion 15 times HCW. As with RPYD, RPWT could be e s t i m a t e d m o r e p r e c i s e l y ( R 2 = . 9 5 v s . 9 0 ) f r o m i m a g analysis data than from yield grade (Figure 3). e
Longissimus Area. Traditionally, carcasses are ribbed between the 12th and 13th ribs following the natural curvature of the ribs. In practice, this leads to a large amount of variation in the angle at which the longissimus is transected, which, in turn, may lead to erroneous estimation of longissimus area. For tender- ness classification, the angle (90) at which the longissimus is transected must be controlled tightly (Shackelford et al., 1997a). Thus, we hypothesized that longissimus area would be more indicative of variation in carcass muscularity if longissimus area was measured on the steak removed for tenderness classification/image analysis rather than if longissi- mus area was measured on the conventionally ribbed
side. Steak longissimus area was t o r e t a i l p r o d u c t y i e l d ( R 2 = . 3 7 v s
more highly related .27) than was CLA.
Thus, when evaluating the predict longissimus area, than CLA.
ability of image analysis to we predicted SLA rather
The single image analysis variable that accounted for the greatest proportion of variation in SLA was
= .85). The best
(Equation 23) RSD and most the validation
optimized R precisely (R data set. 2 2
( three-variable equation . 8 9 ) , C p s t a t i s t i c , a n d = .88) predicted SLA in
Recently, some packers have begun to discount carcasses with extremely small or extremely large longissimus areas. However, it has been difficult for packers to accurately apply these discounts because of the difficulty in subjectively estimating longissimus area when carcasses are evaluated at rates of up to 400 carcasses per hour. Thus, the industry has sought an objective measure of longissimus area. The present image analysis system was quite accurate (Figure 4) at predicting whether a given longissimus area was within the longissimus area target of 71 to 90 cm identified by Tatum (1992). For carcasses with p r e d i c t e d l o n g i s s i m u s a r e a s o f l e s s t h a n 7 1 c m 2 , t h range in observed longissimus area was 58.7 to 76.8 e c m 2 . F o r c a r c a s s e s w i t h p r e d i c t e d l o n g i s s i m u s a r e a s w i t h i n t h e r a n g e o f 7 1 t o 9 0 c m 2 , t h e r a n g e i n o b s e r v e d l o n g i s s i m u s a r e a w a s 7 1 . 6 t o 9 4 . 2 c m 2 carcasses with predicted longissimus areas greater . F o r t h a n 9 0 c m 2 , t h e r a n g e i n o b s e r v e d l o n g i s s i m u s a r e a w a s 8 7 . 1 t o 1 0 9 2 . 7 c m 2 .
Subprimal Cut Weights. Because most beef car- casses are merchandised as boxed-beef subprimals and the subprimal yield of individual carcasses is usually not determined, the true value of most beef carcasses is never known. Thus, technology to measure or predict weights of individual subprimals would allow the beef industry to more accurately estimate true carcass value. To determine whether image analysis could be used to predict the weights of individual subprimals, we regressed predicted RPYD (Equation 15) and HCW against individual weights of each subprimal. Hot carcass weight, by itself, accounted for 23% (cube steak) to 74% (chuck roll) of the variation