Non-destructive, objective methods of determining LMY that are cost effective and can operate in on-farm commercial environments are essential to enhance productivity and profitability of meat supply chains. Two on-farm technologies have been developed reaching varying levels of commercialisation. A hand-held microwave system has shown good capacity for measuring fat depth in both beef and lamb and has been tested in commercial environments by commercial partners. Similarly, a 3-D imaging system for live cattle has been installed in a marshalling race of a beef feedlot demonstrating successful prediction of carcase attributes
Read more here.
KPI 1.16 Continued improvement of the MiS operating system and algorithm
Portable ultra-wide band microwave system coupled with an open-ended coaxial probe(OCP) or Antipodal Vivaldi Antenna (AVA) was tested as non-invasive objective measurement to predict beef carcase single site fat depth at commercial abattoirs. Hot carcase P8 was measured using an OCP (n=435) across two slaughter groups and by an AVA (n=241) across 4 groups. Cold carcase rib fat was measured using AVA (n=598) across 5 groups. A machine learning stacking ensemble technique was used to create the prediction equations. Datasets were grouped by prediction trait (P8 or ribfat) and probe/antenna then randomly divided into 5 groups based on tissue depth. The precision was greatest using the OCP to predict P8 site with an RMSEP of 2.47 mm and R2 of 0.70. The VPA precision was greatest in the hot carcase on the P8 site with an RMSEP of 2.86 mm and R2 of 0.58 compared to the cold carcase rib fat RMSEP of 2.60 mm and R2 of 0.55.
Read more here.
KPI 2.6 Validation of a prototype microwave device to measure fat depth at the rib and P8 sites on live cattle and validate against the corresponding ultrasound and abattoir measurements
Two experiments were performed to test the ability of a portable microwave system coupled with Vivaldi Patch Antenna (VPA) to objectively measure live cattle subcutaneous fatness to predict corresponding carcase traits. Experiment One was performed on-farm, where commercial feedlot slaughter cattle (n=517) were microwave scanned at the P8 (fat depth on
the rump) and rib fat site (fat depth over the m. longissimus, between rib 12 & 13). Corresponding ultrasound measurements were taken (n=315) at the same time as microwave scanning. A machine learning stacking ensemble method was used to create the microwave prediction equations. Datasets were grouped by prediction trait (P8 or rib fat) and randomly divided into 5 groups based on tissue depth. Live animal microwave scanning had greater precision than ultrasound at predicting carcase P8 and rib fat depth. At the P8 site the average RMSEP was 2.61 mm, R2 0.61, bias 0.179 and slope 0.07 mm, and at the rib fat site the average RMSEP was 2.16 mm, R2 0.60, bias 0.301 mm and slope 0.10.
Experiment Two was performed in the abattoir, where commercial slaughter cattle were microwave scanned at the rib fat site immediately prior to entering the knocking box. Microwave scanning in the abattoir had poor prediction of corresponding carcase trait with an R2 of 0.02, RMSEP 2.24, bias 0.019 and slope 0.757.
Read more here.
KPI 2.7 Ability of live microwave scanning in sheep to predict whole body fat composition as determined by live DEXA scanning
This experiment evaluated the ability of a portable microwave system (MiS) coupled with Vivaldi Patch antenna used in mature ewes (n=835) to predict whole body fatness as determined by dual energy x-ray absorptiometry (DEXA). MiS scanning was performed at the C-site (45 mm from spine midline over the 12th rib), Point A3 (20 mm cranial to C-site), the GR-site (110 mm from spine midline over the 12th rib), and a combination of 8 points across the right thorax. Precision of prediction was greatest at the 8 combined points with an average R2 of 0.51 and RMSEP of 1.85 mm, however there was negligible difference in prediction when scanning at only the GR-site with an average R2 of 0.49 and RMSEP of 1.84 mm.
Read more here.
KPI 2.7 Report on antenna and probe design fro portable microwave system
Experiment One details the design and computer simulation of three different probe/antenna performance to penetrate fat and muscle. The Vivaldi Patch Antenna (VPA) had the widest radiation pattern, where as the open-ended coaxial probe (OCP) has a directional radiation pattern. The periodic log antenna (PLA) demonstrated a high gain at a low frequency, however at higher frequencies the gain unexpectedly decreased making it unsuitable for deep tissue measurements.
Experiment Two details the commercial comparison of the VPA, OCP and PLA on sheep and beef carcases. Predicting beef carcase P8 fat depth, the PLA demonstrated the greatest precision of prediction with an average RMSEP of 1.29 mm and R2 of 0.83. Predicting lamb carcase C-site and GR fat depth the VPA and PLA had similar prediction performance where the OCP had a lower precision of prediction. In Experiment B the VPA C-site prediction had
an average RMSEP of 0.94 mm and R2 of 0.67 where as the OCP had an average RMSEP of 1.04 mm and R2 or 0.56. Experiment B VPA GR site prediction had an average RMSEP of 2.80 mm and R2 of 0.67 vs the OCP average RMSEP of 3.31 mm and R2 of 0.56. Experiment C VPA C-site prediction was similar to Experiment B, with an average RMSEP of
1.02 mm and R2 of 0.66. Experiment C PLA average RMSEP was 1.00 mm and R2 0.67. However Experiment C VPA GR tissue depth prediction was less than Experiment B with an average RMSEP of 3.57 mm and R2 of 0.55. Experiment C PLA GR site prediction was similar to VPA with an average RMSEP of 3.70 mm and R2 of 0.50.
Read more here.
Non-destructive, objective methods of determining LMY that are cost effective and canoperate at commercial line speeds are essential to enhance productivity and profitability of red meat supply chains.
Read more here.
KPI 3.11 AUS-MEAT accredited lamb DEXA algorithm performance
An accredited algorithm for the prediction of lamb carcass composition was submitted and accepted by AUSMEAT in September 2022, a report of which can be found as technical report 3.11 – Lamb DXA accreditation submission to AMILSC on behalf of Scott Automation and Robotics and MLA. That technical report shows the results of the algorithm’s predictions, and what bands of weight and composition the algorithm is accredited for.
This technical report details the algorithm that was used, and the method of training and calibration using the Scott Automation and Robotics phantom.
Read more here.
KPI 3.11 DEXA Bone detection algorithm for lamb carcases
The accurate detection of bone in a DXA image is crucial to the overall precision and accuracy of its prediction of overall composition, as the non-bone containing pixels are evaluated independent of the bone-containing pixels. Previous algorithms have relied on thresholding bone containing pixels from non-bone containing pixels at the mean R-value of the whole carcass, with those pixels above the mean being classed as bone
containing. While this was somewhat effective, as it emulated the fundamental equations used to quantify bone content of bone containing pixels, it was unable to detect bone edges, and as such could not use an accurate R-value for soft tissue. This became problematic for carcasses that did not have bone content close to most of the population, such as certain breeds with naturally higher bone content (Merinos), or sheep with a lower amount of soft tissue – usually mutton or diseased sheep. An improved algorithm was proposed and has been accredited as part of the larger DXA
accreditation. This algorithm introduced the evaluation of thickness with the R-value calculation when thresholding for bone content. The implementation of this algorithm sees a very large increase in the precision and accuracy of the prediction of CT bone %, from an R2=0.496 to R2=0.925. There are also modest increases in the precision and accuracy of the predictions of CT fat % and CT lean %. Further improvements on the bone detection algorithm will look to machine learning to detect bone edges and dimensions, a process that is otherwise too slow and difficult at the current chain speed.
Read more here.
KPI 3.11 Initial analysis of lamb DEXA predicting GR tissue depth
Recently, a dual energy X-ray absorptiometry (DEXA) system has been successfully developed to measure lamb lean meat yield (LMY) with high precision and accuracy at abattoir line speed. However, the GR tissue depth (thickness of tissue over the 12th rib, 110 mm from the midline) and fat scores remain the Australian industry standard for estimating fatness in lamb carcases. To satisfy AUS-MEAT measurement requirements, DEXA must be able to predict GR tissue depth in lamb carcases within ±2 mm of the score boundary, with a maximum error of 10%. Lambs (n = 189) with a diverse range in carcase weight and fatness were slaughtered at a commercial abattoir. GR tissue depth and hot carcass weight were measured immediately post-slaughter, before the carcases were then chilled and DEXA scanned the following day. Predicting GR tissue depth using DEXA image components data (pixel number and mean negative log of low energy pixel values) along with mean DEXA values did meet AUS-MEAT requirements for 90% accuracy in fat score prediction. This means the lamb DEXA system could predict single site GR tissue depth to current commercial standards. These findings suggest processing plants could use the objective DEXA system which operates at line speed instead of the slow and subjective palpation or GR tissue measure for fat scoring carcases.
Read more here.
KPI 3.11 Lamb DEXA accreditation submission to AMILSC on behalf of Scott Automation and Robotics
This report has been prepared by ALMTech on behalf of the manufacturer of the lamb DXA device (Scott Automation and Robotics) for the accreditation for predicting CT Fat%, CT Lean% and CT Bone% of sheep carcases. The accreditation trial was conducted at WAMMCO, Katanning, using lamb carcases collected over multiple kill groups from February 2022 through to July 2022 (n=338), with a further group of high weight carcases collected
between August 2022 and December 2022 (n=139).
The experimental and analytical procedure used to assess the repeatability and accuracy performance of the DXA device has been described. The performance of the DXA device was compared against version 1 of the AMILSC approved guidelines for experiments to achieve accreditation of technologies for predicting fat%, lean%, and bone% in sheep carcases (“A carcase composition trait for sheep meat grading technologies” presented to AMILSC on the 17/2/2022). As total carcase composition is loosely associated with hot carcase weight, the accreditation requirements for DXA are tested within three weight categories: light (<22kg); medium (22-28kg); and heavy (>28kg). This creates nine discrete groups within which accreditation tests are applied.
We recommend accreditation of DXA for predicting composition in carcases <22kg with CT Fat % between 10.9% and 30.3%, CT Lean% between 53.2% and 65.0%, and CT Bone % between 14.9% and 25.0%. Additionally, we recommend accreditation of DXA for predicting composition in carcases between 22kg and 28kg with CT Fat % between 14.9% and 35.0%, CT Lean % between 50.9% and 66.2%, and CT Bone % between 13.3% and 18.0%. Finally,we recommend accreditation of DXA for predicting composition in carcases >28kg with CT Fat % between 22.0% and 37.1%, CT Lean % between 49% and 60.6%, and CT Bone % between 11.6% and 17.5%.
Read more here.
KPI 3.11 Standardised methodology for sampling & image analysis, & validation resource data for CT as the calibrating standard for LMY measurement
To support computed tomography (CT) as the reference method for measuring lean meat yield within the Australian beef and lamb industry, it is important to understand the robustness of this measure, and the factors that can influence its estimate of carcase composition. This work was undertaken through a series of experiments that established the immediate repeatability of CT estimates of carcase composition, the impact of changes to
scanning methodology such as carcase sectioning and freeze/thaw protocols, and the effect of machine scanning voltage and CT scan slice width. This work demonstrated several outcomes:
1. CT estimates of carcase composition are almost perfectly repeatable.
2. Carcase sectioning has a small impact on CT estimates of carcase composition, so we propose a standardized carcase sectioning method for lamb to be used when CT is used as the reference standard for calibrating other measurement technologies.
3. CT scan slice width has a small impact on CT estimates of carcase composition, so we propose 5mm slice widths as the standard method for lamb to be used when CT is used as the reference standard for calibrating other measurement technologies.
4. CT scan voltage has a substantial impact on CT estimates of carcase composition, so we propose that a standardized voltage of 120kV is applied for beef when CT is used as the reference standard for calibrating other measurement technologies. Scans captured at other voltages can be corrected to give the 120kV equivalent estimate of carcase composition, but for consistency this should be avoided for calibration purposes.
5. Scanning beef butts in a frozen state decreases tissue Hounsfield unit values, particularly for fat and lean tissues, resulting in substantial variation in their estimated composition within beef butts. Alternatively, scanning these same sections after they have thawed produces values very similar to those scanned fresh, implying that carcases can be frozen and then defrosted prior to CT scanning and still deliver consistent results.
6. When the CT methodology is standardised, it demonstrates substantially better repeatability than determining carcase composition using chemical methods.
7. CT scanning plastic phantoms has demonstrated that the values reported across a range of plastics of different densities only vary slightly between scans and machines. These small variations are likely to be readily accounted for through scanning common calibrating phantoms like the XTE-CT test piece.
Read more here.
KPI 3.11 WAMMCO DEXA calibration
The recent installation of the DEXA scanner at WAMMCO, Katanning, required the calibration of the system against computed tomography (CT), as has been undertaken at all previous DEXA sites around Australia.
100 lambs were selected from the MLA resource flock based at the Katanning research station, with a wide range of phenotypes and genotypes. These 100 lambs were slaughtered and scanned in the new DEXA apparatus as ‘hot’ carcasses, prior to entering a chiller. These 100 carcasses were CT scanned at Murdoch University as the gold standard carcass measurement.
The precision and accuracy measurements for fat, lean and bone were similar to previous DEXA sites after the algorithm adjustment, which was successfully conducted on the plastic phantoms supplied by Murdoch University. This adjusted algorithm produced results that are in line with the expected range of lambs through the scanner, with minimal outliers.
Read more here.
KPI 3.12 Bridging of beef DEXA images to provide LMY feedback
KPI 3.12 DEXA prediction of lean beef trim
This experiment assessed the ability of a commercial-prototype dual energy x-ray absorptiometer (DEXA) to predict the composition of beef sides scanned at abattoir line-speed and thereby determine the lean trim weights from each beef carcase side. It was hypothesised that DEXA provided a better objective measure of whole carcase composition than the current industry standard of P8 fatness, and therefore, would predict lean trim
weights with higher precision and accuracy than P8 fatness. 250 beef carcases representing a wide range in carcase weight and fatness (measured by P8 fat depth) were selected for DEXA scanning and a comprehensive bone-out into retail cuts of meat and trim. Boners assigned lean trim to one of three categories: 60, 85 or 90% chemical lean (CL), based on the visual fatness of the trim according to standard processor protocols. The weight of this trim was predicted in general linear models using hot side weight (HSW) and P8 fatness or DEXA variables as covariates. To address the anticipated difficulty boners faced in visually differentiating 85 and 90% CL, we also used the above models to predict summed 85 and 90% CL trim. In line with our hypothesis, DEXA variables predicted beef trim weights with better precision than P8 fatness, though the differences were small. The precision of prediction varied between trim categories. In the fatter trim category (65% CL), DEXA variables predicted the weight of trim with less precision, with an R-squared of 0.78 and a root mean square error of prediction (RMSEP) of 106 grams. In contrast, precision was improved when predicting the weight of the combined leaner trim category (85+90% CL),
with an R-squared value of 0.92 and an RMSEP of 60 grams. The DEXA system differentiated a greater range in trim weight than P8 fatness and therefore has potential as a novel technology to predict beef trim weights in a commercial setting.
Read more here.
KPI 3.12 Potential for applying DEXA prediction as feedback at the Teys Rockhampton abattoir
KPI 3.13 The value of precise cut weight prediction in the optimisation of lamb carcase processing
The Australian beef industry is focused on improving the objective measurement of lambcarcase yield to improve the valuation of lamb carcases. However, the value that technologies such as DEXA, that have improved the precision of cut weight prediction, represent to plants in the processing of carcases needs clarity. This project thereby assesses the value of using DEXA cut weight predictions in a processing scenario in the lamb Optimisation tool compared to using hot carcase weight (HCWT) and GR cut weight predictions. 191 lambs representing a wide range in HCWT and fatness were measured for GR, DEXA scanned and comprehensively boned out to weigh a wide selection of retail cuts and associated lean trim
and fat. A scenario was then developed in the Carcase Optimisation Tool with 2 to 3 cut options available for each section of the carcase, with cut weight thresholds applied limiting the carcases that could be assigned to the different cut options based on their cut weights. The costs of processing the lambs were input and well as the market value of each retail cut available and the associated fat and trim. The cut weights of the lambs were then entered into the optimiser to determine the optimal processing of the lambs. The optimiser scenario was then repeated using HCWT, GR, DEXA and DEXA image component predicted cut weights to determine the most profitable processing of the carcases. The profits and cut allocations output by the optimiser models were then assessed to determine the difference in profit using cut weight predictions compared to the actual cut weights and examining the number of misallocations caused by imprecise cut weight predictions. Using DEXA variables to predict cut weights resulted in optimised profits closer to those achieved using actual cut weights, and les cut option misallocations in the rack, loin and hindquarter sections of lamb carcases.
Read more here.
Non-destructive, objective methods of determining LMY that are cost effective and can operate at commercial line speeds are essential to enhance productivity and profitability of red meat supply chains. While X-ray based technologies such as DEXA offer excellent “direct measurement” of carcase composition, these can be prohibitively expensive for smaller abattoirs.
Therefore, simpler and more cost effect technologies that predict carcase
composition and therefore provide “indirect measurement” are also required. A microwave system designed and fabricated at Murdoch University has demonstrated excellent precision and accuracy in predicting single site fat depth in beef and cattle. The device is portable, simple to use and poses no risk to human operators or food safety. Similarly, a 3D imaging system for beef, designed at the University of Technology Sydney, and an imaging system for pork commercial known as PorkScan Plus offers robust prediction of lean meat yield.
Read more here.
KPI 1.19 3D imaging in beef carcase LMY
This report details the steps in refining the 3D imaging scanning rig design to integrate into beef abattoir operations at chain speed. In order to provide early LMY estimation the rig was positioned at chiller entrance, acquiring hot carcass side 3D reconstructions as they leave the slaughter floor. The 3D digital shape of the carcass is analysed through a curvature descriptor that is associated to LMY values using a non-linear regression model. We posit that segmenting and identifying consistently correspondences between 3D carcasses allow for improved information gathered. Using 152 carcasses that have been 3D reconstructed and combining the curvature descriptor with HCW resulted in LMY estimation with RMSE 4.1% and R2 of 0:54, indicating that HCW provides independent observations to curvature, and assist in estimating LMY. The presented approach needs to be evaluated with more data to increase the confidence in estimated LMY and evaluating portability of model acrossdifferent deployments.
Read more here.
KPI 1.20 Capability of integrated hyperspectral and 3D imaging cameras to determine subcutaneous fat depth and cover
Hyperspectral imaging has the ability to provide information that goes beyond texture and colour of a surface, near and short-wave infrared spectrum penetrate into the subsurface of materials. Combining disparate sensing modalities, hyperspectral and 3D imaging cameras, could allow to reason about the thickness of the layer of fat over the surface of a carcass
(known as subcutaneous fat). We present a learning-based method to estimate subcutaneous fat depth by leveraging hyperspectral data, modelling light sources and 3D surface shape information. In order to develop such a system, subcutaneous fat depth ground truth is recovered from computed tomography (CT) data via a novel systematic approach involving
ray casting and non-rigidly aligning a 3D reconstruction captured by the 3D imaging camera. We utilise a collection of mid-loins supplied by a commercial abattoir to validate that subcutaneous fat is correlated to total fat with R2 0.76. Our method estimates subcutaneous fat over the entire surface of mid-loins (patches of 1.5mm x 1.5mm) with R2 0.73 and RMSE 1.56mm when compared to CT obtained ground truth. This paves a promising new direction
to rapidly estimate information of carcass fat deposits that are complimentary to other LMY technologies which generally estimated whole carcass properties.
Read more here.
KPI 2.17 On-farm and abattoir suitability of microwave scanner system design
This report details the development of a low cost,portable hand-held Microwave System as an objective measurement technology for measuring various traits in cattle and sheep, live animal and carcase.
Read more here.
KPI 2.17 Report on the ability of a microwave device to predict rib fat and P8 fat depth in beef carcases
A portable ultra-wide band microwave system (MiS) was tested as a non-invasive objective measurement to predict beef carcase single site fat depth at commercial abattoirs. Experiment One used a laboratory calibration technique and tested the effectiveness of MiS coupled with either a Vivaldi Patch Antenna (VPA) or an open-ended coaxial probe (OCP). The VPA was
used to predict hot carcase P8 (fat depth on the rump) across 4 slaughter groups (n=241). The VPA was also used to predict cold carcase rib fat (at the quartering site, 75% along the rib eye muscle) across 5 slaughter groups (n=598). The OCP measured hot carcase P8 across two slaughter groups (n=435). A machine learning stacking ensemble method was used to
create the prediction equations. Datasets were grouped by prediction trait (P8 or ribfat) and probe/antenna then randomly divided into 5 groups based on tissue depth. Precision was greatest using OCP to predict P8 fat depth with a RMSEP of 2.47 mm and R2 of 0.70. The VPA precision was similar for the two tissue depths assessed, hot carcase P8 had an average
RMSEP of 2.86 mm and R2 of 0.58 compared to cold carcase rib fat RMSEP of 2.60 mm and R2 of 0.55.
Experiment two tested the ability of a commercial MiS (C-MiS) device coupled to a VPA probe and on-site calibration to predict P8, rib fat depth and eye muscle area (EMA). The best prediction of hot carcase P8 (n=1650) was from C-MiS device 2 with an average RMSEP of 2.775 and R2 0.81. Hot carcases (n=598) C-MiS scanned to predict cold rib fat had an average
RMSEP of 3.712 and R2 of 0.81. Hot carcases (n=598) C-MiS scanned to predict EMA had an average RMSEP of 10.267 and R2 of 0.43.
Read more here.
KPI 3.14 Compare objective carcase measurement technologies for pig classification systems
Carcass classification systems within the pork industry are based on lean meat yield (LMY). Thus, accurate tools to measure the carcass characteristics predicting LMY are essential to underpin grading systems given they inform pricing grids, herd genetics, product segregation and support consistent trade language, which is essential for industry sustainability. This review summarised the accuracy of several commercially available technologies used for prediction of pork carcass composition. Both optical probes and ultrasound-based devices demonstrated comparable accuracy in the prediction of backfat depth, loin muscle depth, lean meat content and saleable meat yield. Ultrasound devices may have the advantage in that they are non-invasive and in the case of the AutoFom can also predict primal LMY values and cut weights. Some variation in predictions was observed across technologies that could be attributed to operator effect, device settings, reference methods used, and variables included in prediction equations. This highlighted the need for consistent testing methodology, and reference standards to train devices against. Rather than manual dissection methods, which are time consuming, destructive and subject to variation, computed tomography (CT) provides a non-invasive, rapid and highly accurate reference method to train and test device performance against. Continued research into objective measurement technologies for LMY in the Australian
pig industry is recommended, given most investigations have been conducted internationally to data.
Read more here.