Objective Primal Measurement OPM - Pack-off Primal Pick and Pack Fundamental Vision and Sensing Evaluation

Summary

​This project investigated the suitability of different sensing technologies to enable pick and pack automation.  Technologies investigated included Hyperspectral imaging, DEXA, Colour/3D camera and MRI.
Hyperspectral imaging (HSI) can be used to obtain spectral and spatial information of an object over the ultraviolet, visible and near-infrared spectral regions (300nm – 2,600nm).  This information can then be used to characterise different materials.  One particular area of interest is the ability to detect contamination.  There are a number of primal defects for which inspection occurs in the pick and pack area.   A number trials were then performed to provide an initial assessment of which pick and pack defects may be detectable using this technology.  It was found that a significant number of the contaminations investigated could be detected by a hyperspectral camera.
Two key processing tasks for striploins and cube rolls include fat trimming and chine removal.  Due to the inability of a human to see subsurface features, these operations are difficult to perform manually with accuracy.  CT imaging can provide 3D information to perform these tasks but is an expensive technology.  This project aimed to investigate the feasibility in using cost-effective dual-energy x-ray absorptiometry (DEXA) to perform these tasks.
Samples of short loin were sourced from an abattoir and DEXA-scanned.  A number of methods were investigated using this information to try and reliably identify the subsurface features of interest required to perform these tasks.  While some promising progress was achieved, it is felt that a 3D x-ray imaging technology will be required to perform fat trimming and chine removal accurately.
A key requirement for pack-off area automation in this area is the identification of primals.  There is a range of sensing technologies available to achieve this, the most cost-effective and simplest being colour imaging and 3D profiling.  A high-level evaluation was carried out to assess what performance is achievable with this technology, and what may be required to cover its limitations.
Colour and 3D data were taken for a large number of naked beef primals.  Preliminary classification algorithms were written to identify six different primals.  These algorithms were then verified with an independent dataset.  The initial results were positive, especially considering only a single computer training iteration and that a truly independent verification set was used.  More complex classification algorithms would result in further improved performance.  It is also important to consider that accuracies will be dependent on the given application.
SCOTT has a business unit which develops high temperature superconductive wires which can be used to construct powerful magnets which don't require cryogenic cooling and can be used for Nuclear Magnetic Resonance (NMR) / Magnetic Resonance Imaging (MRI) applications.  Helium cooling represents a significant cost and maintenance barrier and removing this requirement allows for industrial MRI machines to be manufactured.  
Many offal contaminants are sub-surface and therefore not detectable with colour and hyperspectral imaging.  Some are also quite similar in density and x-ray absorption to surrounding tissue, making detection using x-ray and CT also challenging.  MRI presents an opportunity for filling this gap.  A number of samples of various offal defects for beef and lamb were obtained from an abattoir.  These were transported back to the SCOTT facility and scanned with a cryogen-free MRI machine.  Most were able to be identified clearly given the right MRI acquisition settings.
Recently, DEXA technology has been shown to allow objective measurement of lean meat yield (LMY).  Commercial installations of this technology exist and it is being further rolled out across the country.  One important counter-balance to LMY however is eating quality (EQ) – care must be taken not to focus breeding on LMY at the expense of EQ.  It has been shown in the literature that there is potential for MRI image features to predict shear force measurements in beef samples.  A number of beef samples were scanned with the MRI scanner (24 samples in total).  These samples were then tested for shear force and intramuscular fat content.  It was shown though that there is potential for MRI data to predict tenderness in beef.  It is recommended that more trials be conducted to build upon these findings.  
This project has effectively demonstrated where a number of sensing technologies sit in terms of ability to enable pick and pack automation.
DEXA sensing technology has now also been demonstrated to be able to scan the bone belt to measure lean loss to waste and rendering.  This measurement can be used to provide real time feedback to improve boning room trimming efficiency, and to address incorrect sortation of lean meat products