Ovine Observer

Jayaseelan Marimuthu, Murdoch University, WA

Author: jayaseelan.marimuthu@murdoch.edu.au

 

In the meat industries, overfat carcasses cause significant economic loss due to the labour required for trimming fat, and the waste it represents. Fat is the most variable component, both in its amount and distribution in the carcass, and on this basis the measurement of carcass fat depth is the cornerstone of most carcass classification schemes for beef, lamb and pork worldwide. Fat depth is often measured manually, however this has the disadvantage of being destructive, subjective and time consuming. The ability to estimate fat depth accurately via a non-invasive and non-destructive technique is therefore highly sought after.

 

One solution showing considerable promise for determining carcass fatness is a Microwave System (MiS) using lower power non-ionising electromagnetic waves. Since biological tissues in animals feature a high contrast in skin, fat, muscle and bone at microwave frequencies MiS can accurately evaluate the fat depth and body composition of carcasses and live animals. This article describes the early design and testing of a low-cost portable MiS machine designed for measuring back fat depth in lamb carcasses with the hypothesis that it will provide a reliable estimate of C-site back fat depth measured 5cm from the midline over the 12th rib in lamb carcasses.

Two groups of mixed sex lambs were slaughtered at a commercial abattoir on separate days. Group 1 consisted of 110 lambs with carcass weights ranging between 17kg and 39kg and C-site fat depth ranging between 0.6mm and 8.5mm. Group 2 consisted of 90 lambs with carcass weights ranging between 18kg and 37kg and C-site fat depth ranging between 0.9mm and 10.7mm. C-site fat depth was measured manually using a grade rule. Both groups were scanned at the C-site using a portable MiS at 1, 2, and 4 hours post mortem, with group 2 also scanned at 24 hours.

 

The prototype MiS operated at frequencies of 320 MHz to 6.5 GHz with output power ranging from -30 dBm to +10 dBm. This was coupled with 2 different prototype broadband antennas: (a) Vivaldi Patch Antenna (VPA) and (b) Periodic Log Antenna (PLA).                    

The VPA antenna was used at all-time points except for the 4h reading for group 2, where the PLA antenna was used. C-site fat depth was estimated using partial least squares regression using leave-one-out cross validation. R-square (R2) and root mean squared error (RMSE) are shown as indicators of precision. R2 is a measure of the variation explained by the prediction model, with 1 being a perfect prediction, and RMSE is a measure of the error in the prediction model, with a smaller RMSE indicating the prediction based on the microwave C-site fat depth is close to the actual fat depth.

To further test transportability, the prediction equation for fat depth trained on the 2h post-mortem data in group 2 was used to estimate the C-site fat depth of lamb carcasses from group 1 at 1, 2 and 4h post-mortem and group 2 at 1, 4 and 24h post-mortem.

 

For the relationship between actual versus predicted C-site fat depth, R2 and the RMSE of the prediction are shown as indicators of precision, and slope of the relationship and bias estimates are shown to represent accuracy. Bias represents the difference between the predicted and actual values at the mean of the dataset.

Internal cross-validation demonstrated good precision for the prediction of C-site fat depth using the MiS (Table 1).  This was evident within both groups and across all measurement times, with RMSE values ranging from 0.99mm to 1.34mm. Measurements taken using the PLA antenna at 1h and 2h in group 1.  Within groups, the predicted fat depth values for the same carcasses were highly correlated between time points with a perfect correlation having a value of 1. In group 1, these correlations ranged between 0.85 and 0.91, and in group 2 the correlations ranged between 0.91 and 0.98.

 

Validation testing of the equation derived at 2h in group 2 demonstrated good transportability across the remaining datasets (Table 2) where the VPA antenna was used. Precision estimates were only slightly reduced with RMSE values ranging from 1.17mm to 1.24mm, and slope values (all close to 1) and bias estimates (at worst 1.48mm) indicating good accuracy.

 

The exception to this was for the group 1 data set at 4h where precision was reduced and accuracy was poor. However, this result is not surprising as these measurements were taken using a different antenna (PLA) to that which generated the training data in group 2 (VPA).
 

RMSE = Root mean squared error; VPA = Vivaldi Patch Antenna; PLA = Periodic Log Antenna

 

Figure 3 shows the relationship between microwave and grade rule measurement of C-site fat depth where the microwave prediction at 24h in group 2 was based on the equation derived at 2h in group 2. The residual, or the difference between the two measurements is the distance of the marker from the 45° line (Figure 3). If the residuals are on the black line it indicates the prediction of C-site fat depth is equal to the actual C-site fat depth. The orange trend line for the residuals closely matches the black line indicating good accuracy of the prediction.

 

These results demonstrate the capacity of a portable prototype MiS system to estimate fat depth at the C-site in lamb carcasses non-invasively. Results suggest that this measurement can be taken up to 24h post-mortem with little loss in precision and accuracy, although equations derived for one antenna (VPA) cannot be seamlessly applied to an alternative antenna design (PLA).