HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by McClure, E. K. Articles by Smith, G. C. Search for Related Content PubMed PubMed Citation Articles by McClure, E. K. Articles by Smith, G. C.

Evaluation of the E+V video image analysis system as a predictor of pork carcass meat yield¹

Department of Animal Sciences, Colorado State University, Fort Collins 80523-1171

² Correspondence:
12A Animal Science Bldg. (phone: 970-491-6244; fax: 970-491-0278; E-mail:
jscanga{at}lamar.colostate.edu).

	Abstract

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 
This study was conducted to assess the ability of the VCS2001 (E+V, Oranienburg, Germany) video image analysis system to predict pork carcass composition. Pork carcasses (n = 278) were selected from a commercial packing plant to differ in weight, Fat-O-Meater (FOM) predicted percentage lean, and gender. Carcasses were imaged three times with the VCS2001, chilled overnight, and then sequentially fabricated into boneless subprimals. The VCS2001 accurately predicted the weight of total saleable product (R² = 0.88, root mean square error [RMSE] = 1.84) and fat-corrected lean (R² = 0.92, RMSE = 1.66), but autocorrelation existed between dependent and independent variables. The VCS2001 was acceptably accurate and precise in predicting weights of bone-in ham (R² = 0.83, RMSE = 0.80), bone-in loin (R² = 0.74, RMSE = 1.17), loin lean (R² = 0.77, RMSE = 0.82), belly (R² = 0.78, RMSE = 0.94), sparerib (R² = 0.55, RMSE = 0.28), and boneless shoulder (R² = 0.73, RMSE = 0.79). Weights were more accurately predicted than yields (as a percentage of hot carcass weight) of total saleable product (R² = 0.47, RMSE = 1.97) or total fat-corrected lean (R² = 0.44, RMSE = 1.89) using VCS2002, and it did not accurately predict percentages of bone-in ham (R² = 0.45, RMSE = 1.13), ham lean (R² = 0.32, RMSE = 1.46), bone-in loin (R² = 0.29, RMSE = 1.36), loin lean (R² = 0.56, RMSE = 0.90), belly (R² = 0.43, RMSE = 1.08), sparerib (R² = 0.08, RMSE = 0.32), or boneless shoulder (R² = 0.30, RMSE = 0.88). New prediction models and equations were developed using VCS2001 output variables plus hot carcass weight to predict weight of total saleable product (R² = 0.89, RMSE = 1.72) and fat-corrected lean (R² = 0.93, RMSE = 1.55) with very minimal increases in accuracy and precision over that achieved using E+V-programmed models and equations. Use of new prediction models and equations marginally improved accuracy and precision of estimations of total saleable product yield (R² = 0.56, RMSE = 1.81) and fat-corrected lean yield (R² = 0.57, RMSE = 1.67) over that achieved using E+V-programmed models and equations. The VCS2001 was not able to predict pork carcass composition more accurately than existing technology; therefore, further development is needed to assure commercial viability of this instrument.

Key Words: Carcass Composition • Instrumentation • Lean • Pork

	Introduction

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 
Over 70% of U.S. pigs are marketed through "carcass-merit" pricing systems, in which predicted carcass composition and lean meat yield determine value (NPPC, 2000). Accurate estimates of carcass composition are critical to the success of a value-based marketing program (Boland et al., 1994). Forrest et al. (1989) indicated that the development of computerized, noninvasive systems operating at faster chain speeds would enhance the application of optical probes and other methods that could predict lean weight and would also greatly improve value-based marketing systems.

Several technologies have been evaluated to determine the accuracy and precision for predicting carcass red meat weights and yields (as a percentage of hot carcass weight), including ruler measurements, hand-held optical probes (Fat-O-Meater [FOM]; SFK Technologies, Herlev, Denmark), reflective spectroscopy probes (Hennessy Probe; Hennessy Grading Systems Ltd., Auckland, NZ), automated ultrasound scanning devices (AutoFOM; SFK Technologies, Herlev, Denmark), ultrasonic scanning, bioelectric impedance analysis, total body electromagnetic conductivity (TOBEC), and, more recently, video image analysis. However, use of video image analysis to predict pork carcass composition and subprimal weights and yields has not been extensively researched. Therefore, this study was conducted to evaluate the ability of the VCS2001 video imaging system to predict pork carcass cutability.

	Materials and Methods

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 
Pork carcasses (n = 278) were selected by Colorado State University personnel over a period of 5 wk in a commercial packing plant (chain speed of approximately 1,200 animals/h). Scalded/dehaired carcasses were chosen (Table 1) to reflect variation in gender (gilts and barrows), hot carcass weight (<81.7, 81.8 to 86.2, 86.3 to 90.7, and >90.7 kg), and percentage of carcass lean (<50%, 50 to 52.5%, 52.6 to 55%, 55.1 to 57.5%, and >57.5%) estimated with a FOM (SFK Technologies). Selected carcasses (approximately 25 at a time) were circulated past the VCS2001 at chain speed, side-railed from production, and subsequently circulated past the imaging system twice more for repeatability analyses. The VCS2001 (E+V, Oranienburg, Germany) is a fully automatic carcass evaluation system that incorporates a one-color camera, which images the internal view (body cavity and exposed surface of the vertebral column) of split pork carcasses. Raw output data include linear measurements of carcass dimensions, which are subsequently used in computer algorithms to predict percentage of carcass lean, total lean weight, and weights of lean from the sparerib, bone-in ham, bone-in loin, boneless ham, boneless loin, belly, and shoulder of pork carcasses. Selected carcasses were then railed back onto production rails, "snap-chilled," held overnight, and fabricated.

Each carcass side was weighed and sequentially fabricated off-line into primal and subprimal cuts, according to USDA institutional meat purchasing specifications (IMPS; USDA, 1997), although cuts were not certified by a USDA representative. The fresh ham (IMPS 401, skinned) was removed and then sequentially fabricated into an outside ham (IMPS 402E, denuded), inside ham (IMPS 402F, denuded), knuckle (IMPS 402H, denuded), light butt, dark butt, inside shank, and outside shank. The tenderloin tip and the flank were removed from the ham and collected with the inside shank for fat analysis. The loin (IMPS 410) was sequentially fabricated into a boneless sirloin (IMPS 414A, denuded), boneless loin (denuded), tenderloin (IMPS 415A), back ribs (IMPS 422), and loin riblets (IMPS 424). The belly (IMPS 408) was sequentially fabricated into a skinless belly (IMPS 409, trimmed), fat back (skin removed), spareribs (IMPS 416), premium spareribs, and BBQ rib (IMPS 416C). The shoulder was fabricated into a picnic shoulder (IMPS 405) and then sequentially dissected to generate a picnic cushion (IMPS 405B, denuded), boneless picnic (IMPS 405A, skinned), and Boston butt (IMPS 406, denuded). The neck bones (IMPS 421, trimmings removed), jowl (IMPS 419, skinned, inedible trim removed), front foot (IMPS 420), hind foot (IMPS 420A), and tail were also weighed and recorded following removal. Corresponding skin, fat, lean trimmings, and bone from each sequential fabrication process were removed, weighed, and weights were recorded. Lean trimmings from the belly (50% and 80% lean), loin (50% and 80% lean), and ham (80% lean) were collected for fat analysis. Aggregated weights of all parts were collected from each sequential fabrication step. Only cutout data that summed to at least 97.5% of the initial cold carcass side weight were retained and included in data analyses (all fabricated carcasses met this criterion).

Carcass samples of the IMPS 419; neck bone trimmings; Boston butt neck portion; IMPS 405A; IMPS 405B; 80% loin trimmings; 50% loin trimmings; 80% belly trimmings; 50% belly trimmings; dark butt, tenderloin tip, flank, and inside shank trimmings; and 80% ham trimmings were ground twice using a Hobart grinder (Troy, OH) with a 0.32-cm plate, homogenized with a kitchen food processor, subsampled (approximately 50 g), and stored at 3 to 5°C for further analyses. Subsamples were analyzed to determine percentages of moisture, fat, and protein (AOAC, 1990) using a microwave moisture analyzer (Matthews, NC).

"Total saleable product" was the sum of weights of closely trimmed primals and subprimals, which included the IMPS 402E, IMPS 402F, IMPS 402H, light butt, IMPS 414A, boneless loin (0.32 cm trim), IMPS 415A, IMPS 409 (trimmed), IMPS 405A, IMPS 405B, and IMPS 406 (denuded). "Fat-corrected lean" was the sum of the boneless, denuded, and chemically analyzed lean components of the carcass, including IMPS 402E, IMPS 402F, IMPS 402H, light butt, IMPS 414A, boneless loin, (denuded), IMPS 415A, boneless Boston butt, and the adjusted (100% lean) weights of neck bone trimmings, IMPS 419, IMPS 405A, IMPS 405B, Boston butt neck portion, 80% belly trimmings, 50% belly trimmings, 80% loin trimmings, 50% loin trimmings, 80% ham trimmings, and the inside shank, dark butt, tenderloin tip, and flank trimmings, all from the IMPS 401 ham.

Fabrication weights and percentages were also generated for primals and subprimals. "Bone-in ham" represented the IMPS 401, skinned, 0.64-cm trimmed ham. "Ham lean" was defined as the outside ham (IMPS 402E), inside ham (IMPS 402F), knuckle (IMPS 402H), and light butt, and weights of ham trimmings, dark butt, tenderloin tip, flank, and inside shank, adjusted to 100% lean after fat analysis. "Bone-in loin" represented the IMPS 410, 0.64-cm trimmed loin. "Loin lean" was generated from the tenderloin (IMPS 415A), boneless sirloin (IMPS 414A), boneless denuded loin, and loin trimmings, adjusted to 100% lean after fat analysis. "Belly" was generated from the boneless, skinless belly (IMPS 409) and belly trimmings, which were adjusted to 100% lean after fat analysis. "Sparerib" was generated from the trimmed sparerib, breastbone removed (IMPS 416C). "Shoulder" was generated from the boneless picnic shoulder (IMPS 405A), the picnic cushion (IMPS 405B), neck portion of the Boston butt, all adjusted to 100% lean after fat analysis, in addition to the denuded bone-in Boston butt (IMPS 406).

Statistical Analysis

All statistical analyses, including descriptive statistics, simple correlations, and multiple regression procedures were performed using SAS (SAS Inst., Inc., Cary, NC). Eighty-nine different linear measurements of the carcass were generated using the VCS2001 and multicollinearity among these variables was addressed by evaluating the relationships among the independent VCS2001 output variables. Because several independent variables were correlated (P < 0.05), variables were summed together or simple averages of the correlated independent variables were examined, for inclusion into prediction equations.

Multiple linear regression procedures were used to develop prediction equations for percentages of saleable lean, fat-corrected lean, bone-in ham, ham lean, bone-in loin, loin lean, belly, sparerib, and shoulder. Forward selection, backward elimination, and forward stepwise model selection methods were used to determine which variables were common and significant (P < 0.05) in all three models. Variables not selected by any of the three regression selection methods were omitted from the regression analysis, and the three regression selection methods were performed again with the narrowed pool of independent variables, and models for predicting weights and percentages of total saleable product, fat-corrected lean, bone-in ham, ham lean, bone-in loin, loin lean, belly, sparerib, and shoulder were constructed.

Five regression models with different numbers of independent variables were evaluated to predict weight of subprimal cuts, saleable product, and fat-corrected lean, as well as their corresponding percentage yields, using the highest R² values and the lowest root mean square errors (RMSE), as well as appropriate predicted residual sum of squares (PRESS) and Mallow’s C(p) statistics. The RMSE was calculated as a measure of precision.

Mean absolute difference and standard deviation were calculated to evaluate the instrument’s predictive repeatability. The absolute difference was calculated as the difference between individual estimated carcass lean percentage and the average estimated carcass lean percentage that was generated as carcasses were circulated past VCS2001 three times. The absolute values for all carcasses were added together and averaged to determine the mean absolute difference of readings by the VCS2001. Least squares ANOVA components also were used to assess the repeatability of VCS2001 estimates of carcass traits among pork carcasses (n = 275) using the following model:

where Y_i = estimated percentage of carcass lean of the ith carcass, µ = overall mean of percentage carcass lean, _i = random effect of the ith carcass, and _i = variance explained by the VCS2001.

	Results and Discussion

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 
Descriptive statistics for characteristics of the 278 pork carcasses in the sample population are presented in Table 2. Variation in carcass characteristics for the sample population spanned that of the processing facility’s population and generally represented industry averages for weight (89.8 kg; USDA-AMS, 2002) and backfat thickness (2.79 cm; Cannon et al., 1995).

Weights of bone-in ham (Table 4), ham lean (Table 4), bone-in loin (Table 5), loin lean (Table 5), belly (Table 6), sparerib (Table 6), and boneless shoulder (Table 6) were also predicted more accurately and precisely than was achieved by use of existing VCS2001 equations, and also indicated strong weight-to-weight relationships among the individual subprimal weights as dependent variables and hot carcass weight as an independent variable.

Coefficients of determination for VCS2001 predictions of percentage of total saleable product (R² = 0.47, RMSE = 1.97) and fat-corrected lean (R² = 0.44, RMSE = 1.89) using new regression coefficients were higher by 5 and 8%, respectively, than those produced using current prediction equations (Table 3). Predictions of percentages of carcass weight, rather than subprimal weights, of bone-in ham (Table 4), ham lean (Table 4), bone-in loin (Table 5), loin lean (Table 5), belly (Table 6), sparerib (Table 6), and boneless shoulder (Table 6) were all less accurate, potentially limiting their commercial application. For those reasons, new regression equations were developed to determine if subprimal yields, percentages of total carcass saleable product and fat-corrected lean could be more accurately predicted.

Coefficients of determination (R²), RMSE, and PRESS statistics for regression equations using independent FOM output variables, compared to use of the VCS2001, are presented in Table 3. Accuracy of prediction of carcass weights and percentages using FOM output was comparable to that of using the current VCS2001 regression equations for predicting weight of saleable product and fat-corrected lean because both included hot carcass weight in the prediction equations. Partial R² values, indicate that hot carcass weight accounted for a majority of the variation in prediction of weight endpoints (Table 3). All equations demonstrated that predicted weights were dependent on independent variables that were expressed as a function of hot carcass weight. The percentage of saleable product and fat-corrected lean from a carcass was predicted more accurately (R² = 0.51 vs. 0.42 and 0.54 vs. 0.36, respectively) and with greater precision (RMSE = 1.87 vs. 2.04 and 1.69 vs. 1.99, respectively) by the FOM than by the VCS2001.

Coefficients of determination, RMSE values, and PRESS statistics for regression equations developed by NPPC (2000) to calculate the percentage of standardized fat-free lean of a carcass are presented in Table 3. Because the NPPC (2000) percentage lean equation uses variables similar to those used by the FOM, both predicted percentage of total saleable product and fat-corrected lean with levels of accuracy and precision comparable to those achieved by use of the VCS2001. The FOM, however, still predicted saleable product and fat-corrected lean percentages with higher degrees of accuracy and precision compared to the NPPC (2000) prediction equation, as demonstrated by higher R² values and lower PRESS statistic values.

New regression equations were developed using various selection methods in an attempt to improve the current predictive ability of the VCS2001 using independent output variables and computed variables that were generated by either adding together or averaging similar and/or collinear independent variables (Table 3). When predicting the weight of saleable product, fat-corrected lean, bone-in ham, ham lean (Table 4), bone-in loin, loin lean (Table 5), belly, sparerib, and shoulder (Table 6), new regression equations resulted in minimal improvement in accuracy (R² values) and no improvement in precision (RMSE). Nevertheless, PRESS statistics indicated improvement in efficiency for all previously mentioned dependent variables. However, predicting percentages of bone-in ham, ham lean, bone-in loin, belly, sparerib, and shoulder was still marginally accurate using different independent output variables because the variables used were designed to predict weights of lean from pork carcasses and individual subprimals. Using selected and newly developed independent output variables to predict percentages of saleable product and fat-corrected lean improved the coefficients of determination by 14 and 21%, respectively, compared with the unaltered, existing equations of the VCS2001 (Table 3).

A simulated multi-instrument system that combined estimates by both FOM and VCS2001 to predict saleable product, fat-corrected lean, and subprimal weights and percentages could more accurately predict the same dependent variables than the models developed with new VCS2001 output variables alone (Table 3). Coefficients of determination (R²) for percentages of saleable product (Table 3), fat-corrected lean (Table 3), and loin lean (Table 5) in comparison to R² values for the newly regressed equations increased by 0.05, 0.06, and 0.08, respectively, demonstrating that if the FOM output could be combined with VCS2001 output, the percentage of carcass lean could be predicted more accurately. Prediction equations for weights of saleable product and fat-corrected lean did not improve substantially in comparison to those for percentages of saleable product, fat-corrected lean and subprimal weights because autocorrelation between hot carcass weight and the two dependent variables was evident in the prediction equations.

Two studies of video image analysis systems to predict saleable meat yield and fabrication yields of wholesale cuts reported R² and RMSE values (Cannell et al., 2002; Brady et al., 2003) comparable to those in Table 3 for percentage of saleable product. Brady et al. (2003) predicted boneless and bone-in saleable meat yield (as a percentage of chilled side weight) in lamb carcasses with coefficients of determination of 0.63 (RMSE = 0.028) and 0.62 (RMSE = 0.021), respectively. The best LVS equation included hot carcass weight and six independent LVS output variables (Brady et al., 2003). Cannell et al. (2002) reported a coefficient of determination of 0.64 (RMSE = 1.41) when predicting actual fabrication yields of beef wholesale cuts (as a percentage of chilled side weight) using a commercial video image analysis system (Computer Vision System, CVS). The best equation included the independent variables of three hot-camera linear carcass dimensions, hot carcass weight, CVS-measured fat thickness at the midpoint of the ribeye, and CVS-measured ribeye area.

Repeatability of VCS2001 estimates of percentage of carcass lean, measured via calculated mean absolute difference, demonstrated that between repeated measurements of each carcass, there was a 0.46% difference (SD ± 0.48%) in average estimated percentage lean. This system therefore would produce carcass lean estimates that are, on average, 0.46% different from the average predicted percent carcass lean, and as much as 1.42% (± 2 SD) from the average predicted carcass lean 95% of the time. Comparatively, many pricing grids segregate carcasses into value groups with intervals of 0.5 to 1.5% lean.

Using the least squares ANOVA components for repeatability, differences among carcasses were responsible for 91.85% of the observed variability in the percentage of carcass lean in the sample population and 8.15% of the variation was explained by variability in VCS2001 readings. Thus, in this study, the VCS2001 had an error rate of 8.15%. Presentation was undoubtedly a factor in this error rate because some carcasses exhibited bad splits (i.e., split unevenly), carcasses swayed while passing the camera, or some had missing parts (removed due to contamination), which would have caused miscalculations in the estimated percentage carcass lean.

	Implications

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 
The VCS2001 can be used to predict weights of saleable product, fat-corrected lean, bone-in ham, bone-in loin, loin lean, and belly with high levels of accuracy when hot carcass weight is included as a predictive factor. The VCS2001 is similar to Fat-O-Meater for predicting the weight of total saleable product and fat-corrected lean. In industry, combining the use of VCS2001 with that of Fat-O-Meater would be advantageous because Fat-O-Meater could provide an estimate of percentage of carcass lean on those carcasses for which the VCS2001 cannot obtain appropriate visual images (i.e., due to bad splits and other slaughter/dressing defects). In this study, VCS2001 was not sufficiently repeatable in predicting saleable product yields to be used as a value-determining tool and further development is needed to assure commercial viability of this instrument.

	Footnotes

 
¹ The authors wish to thank Swift and Co. for their assistance in conducting this project.

Received for publication October 2, 2002. Accepted for publication January 28, 2003.

	Literature Cited

Top Abstract Introduction Materials and Methods Results and Discussion Implications Literature Cited

 


AMSA. 2001. Meat Evaluation Handbook. Am. Meat Sci. Assoc., Savoy, IL.

AOAC. 1990. Official Methods of Analysis. 15th ed. Assoc. of Offic. Anal. Chem., Arlington, VA.

Boland, M. A., K. A. Foster, J. T. Akridge, and J. C. Forrest. 1994. Economic analysis of using electromagnetic scanning in a packing plant. Pages 69–72 in Purdue Univ. Swine Day Rep., West Lafayette, IN.

Brady, A. S., K. E. Belk, S. B. LeValley, J. A. Scanga, J. D. Tatum, and G. C. Smith. 2003. An evaluation of the Lamb Vision System as a predictor of lamb carcass red meat yield percentage. Sheep Goat Res. J. (In press).

Cannell, R. C., K. E. Belk, J. D. Tatum, J. W. Wise, P. L. Chapman, J. A. Scanga, and G. C. Smith. 2002. Online evaluation of a commercial video image analysis system (Computer Vision System) to predict beef carcass red meat yield and for augmenting the assignment of USDA yield grades. J. Anim. Sci. 80:1195–1201.[Abstract/Free Full Text]

Forrest, J. C., C. H. Kuei, M. W. Orcutt, A. P. Schinckel, J. R. Stouffer, and M. D. Judge. 1989. A review of potential new methods of on-line pork carcass evaluation. J. Anim. Sci. 67:2164–2170.[Abstract/Free Full Text]

Higbie, A. D., T. D. Bidner, J. O. Matthews, L. L. Southern, T. G. Page, M. A. Persica, M. B. Sanders, and C. J. Monlezun. 2002. Prediction of swine carcass composition by total body electrical conductivity (TOBEC). J. Anim. Sci. 80:113–122.[Abstract/Free Full Text]

Liu, Y., and J. R. Stouffer. 1995. Pork carcass evaluation with an automated and computerized ultrasonic system. J. Anim. Sci. 73:29–38.[Abstract]

NPPC. 2000. Pork Composition and Quality Assessment Measures. Natl. Pork Prod. Council, Des Moines, IA.

USDA. 1985. United States Standards for Grades of Pork Carcasses. USDA-AMS Livestock and Seed Division, Washington, DC.

USDA. 1997. Institutional Meat Purchasing Specifications: Series-400, Pork. USDA-AMS Livestock and Seed Division, Washington, DC.

USDA. 2002. National Daily Hog and Pork Summary. December 31, 2002. USDA-AMS Livestock and Seed Division, Washington, DC.

This article has been cited by other articles:

				A. P. Schinckel, M. E. Einstein, K. Foster, and B. A. Craig Evaluation of the impact of errors in the measurement of backfat depth on the prediction of fat-free lean mass J Anim Sci, August 1, 2007; 85(8): 2031 - 2042. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by McClure, E. K. Articles by Smith, G. C. Search for Related Content PubMed PubMed Citation Articles by McClure, E. K. Articles by Smith, G. C.