1. Introduction
In the last decades, studies related to meat characteristics and carcass quality of lambs have been carried out using traditional instrumental methods, such as chemical and physical analyses [
1,
2]. Physical analyses by complete dissection are the most widely used ways of determining carcass composition. However, the carcass dissection is a time-consuming and expensive process, requiring skilled labor and depreciating the carcass, and is also associated with inconsistency and instability [
3,
4].
The lack of simple, non-destructive, rapid and reliable methods to assess carcass classification and the characteristics of carcass joints has been one of the barriers to developing quality control systems in the meat industry [
5,
6]. To overcome these difficulties, several efforts have been employed to develop fast, simple, objective and inexpensive methods of establishing measurements of the carcass and its tissues or cuts [
7,
8,
9].
The research and development of non-destructive, non-invasive technologies have mainly been driven by the need for objective and accurate carcass trait selection, to improve carcass grading or to assess meat quality that satisfies consumer demands, thus add fairness in determining the carcass value and reduce labor requirements for processors [
4,
10,
11]. Video image analysis (VIA) is an example of such technology; it has been widely researched for cattle [
3,
12,
13] and sheep, although to a lesser extent for the latter [
7,
14]. According to Scholz et al. [
6] and Ngo et al. [
14], the emphasis on the use of VIA is to imitate visual evaluation, however, in an objective way. The latter authors [
14] present a flexible, low-cost and objective image analysis system applied in slaughterhouses, helping determine the cuts and lean prediction weight of lamb carcasses. Usually, the works published on VIA apply to sheep use as far as possible. The carcass spectrum of lambs slaughtered in different production systems typically ranges in weight between 15 and 30 kg [
7,
8,
14]. However, there is a lack of information for light carcasses, and in that regard, this study aims to evaluate the accuracy of a flexible, low-cost VIA system in predicting the weight and yield of lean, commercial cuts from light lamb carcasses.
4. Discussion
The cuts of the groups separated by commercial value presented in this work follow the specifications of light lambs in the Northern region of Portugal [
23] and the southern European countries [
24,
25,
26]. The average weight of the cold carcass was 4.52 kg; that is according to Borrego Terrincho–PDO specifications [
23]. The value of the coefficient of variation of the carcass weight reflects the variability of consumer preferences.
Measurements of length, area, width and angle made through VIA systems were previously reported in other studies [
14,
17,
18,
27].
The total number of VIA measurements used overall was quite large, even after excluding the poorest predictors, but this was due to the variation in the predictors included in the final models for the different cut groups. A similar variation was recently reported by Gardner et al. [
28] in their study using dual-energy X-ray absorptiometry to estimate commercial cut weights. Only the models to estimate cut weights and lean meat weights with CCW excluded from the analysis tended to use the same predictors in the present study. Even in this case, the models for estimating LVC (both in weight and lean meat weight) show some variation.
The present results confirm that carcass weight is the most significant variable in the estimation models for cut weights [
14,
29]. Brady et al. [
29], including carcass weight in their models, had already obtained high accuracies in the prediction of cut weights, explaining 72.2% to 85.8% of the variation of the shoulder, rack, loin and leg cuts. The accuracy of the models now developed is in line with the results of Ngo et al. [
14], who, using measurements obtained with a lamb digital grading system, in addition to carcass weight, obtained models that explained 94% to 95% of the variation of primal cut weights. Working with goats, Monteiro et al. [
30] also showed an accuracy of above 99% for their predictive model of prime cuts weights, with HCW, by itself, explaining 99% of the variation observed. Rius-Vilarrasa et al. [
31] had already reported a high value of models using only VIA measurements to predict cut weights (0.86 < R
2 < 0.97 for several primal cuts and R
2 = 0.99 for total primal cuts). Although showing lower predictive value than the models obtained by Rius-Vilarrasa et al. [
31], the present results confirmed that VIA measurements, by themselves, are good predictors of cut weights, and their value is not limited to providing complementary information to increase the prediction value of models including CCW as the independent variable.
The poor accuracy of the present models in predicting cut percentages confirms the results obtained by Monteiro et al. [
30], with a model based only on one VIA measurement that explained 19.6% of the variation observed in the percentage of primal cuts in goats. However, these results contrast with the high accuracies shown by Brady et al. [
29] and Cunha et al. [
32], including HCW (and also, in the case of Cunha et al. [
32], the longissimus muscle area) in their prediction models. These two studies explained, respectively, 57.9% and 64.1% of the variation of subprimal cut percentages.
Although carcass weight is the most important factor determining the weight of the different cuts, there can be substantial variation in the weight of saleable cuts obtained from carcasses of similar weights, mainly due to variation in fatness, as pointed out by Gardner et al. [
28]. The present results indicate that VIA measurements can increase the accuracy of lean meat weight estimates for different cut groups when used as predictors together with CCW and, just by themselves, can provide such estimates with high accuracy.
Concerning lean meat percentage, Normand and Ferrand [
33] had already shown little effect of carcass weight, while Stanford et al. [
34] showed carcass weight as the main predictor of saleable meat percentage together with VIA measurements. The present results were not clear about this subject, showing CCW as the first predictor to include in models for predicting lean meat percentage in LM_MVC and LM_AllC, but excluding CCW from the models for prediction of the same trait in LM_HVC and LM_LVC. The 36.4% variation explained in the present study for lean meat percentage in LM_AllC, without CCW in the predictive model, is significantly smaller than the 45% variation in lean meat percentage explained by Normand and Ferrand [
35] using only VIA predictors. Also, without carcass weight as a predictor, Einarsson et al.’s [
27] models explained 60%, 31% and 45% (model 1), and 57%, 30% and 47% (model 2) of the variation observed, respectively, for lean meat percentages of the leg, loin and shoulder. With a model using carcass weight and several VIA measurements, Stanford et al. [
34] explained 71% and 62% of the variation observed among lean meat percentages in the leg and shoulder cuts. Including HCW and the longissimus muscle area in their prediction models, Cunha et al. [
32] explained 67.8% of the saleable meat yield percentage variation. Except for Einarsson et al.’s [
27] models for the loin and shoulder that showed a moderate predictive value of lean meat percentage, these other three studies have shown models with significantly higher predictive values for this trait than the present models.
All models obtained were good predictors (RPD > 2.0) of cut weights and lean mean weights. The ones including CCW, in addition to VIA measurements, were even excellent predictors (RPD > 2.5), as well as the ones without CCW, in the case of MVC and AllC. However, with RPD values between 1.02 and 1.29, the models now obtained showed poor value for predicting cut percentage and lean meat percentage.