Preliminary examination of the ability of x-ray or dual-energy X-ray of skulls to predict animal age and automate dentition checks.

Canada’s beef industry has conducted Beef Quality Audits every few years starting in the mid-1990’s. The Audits provide very valuable information about what our industry is doing well in terms of producing quality beef carcasses (e.g. marbling, tenderness), and where it can do better in terms of avoiding some of the estimated $200 million annual cost of quality defects such as bruising, tag, liver condemnations and injection site lesions. However, doing these audits on a five-year basis means that we are often learning about problems that started quite a while ago. However, new technologies may provide the opportunity to track carcass and offal quality (and defects) on an ongoing basis and allow us to recognize and respond to emerging quality challenges and opportunities in a timely manner. This team investigated whether x-ray images of the skull can predict animal age at the slaughter plant. They also further developed grading and photographic imaging to automatically collect hide color, tag, horn, brand, bruise, injection site lesion, muscling, hump height, liver abscess, and age data. They added new programs evaluating marbling texture, blood splash, dark cutter, yellow fat and tenderness into the e+v ribeye camera. These developed systems will allow the information they gather to flow back to feedlots. Current iDXA equipment does not provide sufficient quality images for age identification. As low energy systems would be preferred in a plant environment, there could be opportunity to develop a task specific dual energy system capable of providing higher quality images of cow skulls.
A portable x-ray camera could be utilized to achieve quality images for automatic image analyses and age verification within a packing plant environment. However, a mechanism to ensure repeatable placement of the camera in relation to the skull needs to be developed.
Accurate CNN models for age classification are possible; in a plant environment, continuous improvement of the models can be achieved as the number of images increases using an ongoing learning process.
This may open strategic marketing opportunities based on accurate age range differentiation of beef carcass. The present results provides the background knowledge, thresholds or algorithms, to automatically segregate carcasses based on age range in real time using CNN procedures on mandible x-ray images and conventional visible images. This information might be the foundation for further progress towards automation of on-line procedures, and development of learning software algorithms such that algorithms can be improved automatically as additional data is collected.