Justin Sexten: Mass Photography
There are few measurements used in cattle management of greater importance than weight. Certainly one can argue the categorical variables of pregnancy status and death loss are more impactful to profitability due to the challenge of marketing the absence of productivity.
When we consider the influential variables we measure and manage to, weight tops the list as weight-based decisions are at every stage of the beef supply chain, birth to carcass. On every side of the transaction, purchase, sales and inputs. Even the ability to influence weight has value, in both positive and negative directions: increased gain and ability to reduce mature size.
Once you get past the bale and bucket as units of measure, feed inputs are weight driven as well. We’ll save the review of technology available to measure feed and forage inputs for another day.
Zhuoyi Wang and co-workers at the University of Guelph discussed the technology evolution used to predict body weight. To clarify, this article was not a review of the move from a balance, to bar and ultimately digital scale, rather a review of imaging technology to predict weight without having to capture or constrain the animal.
Technology enabled weight estimation systems address the age old challenge of weight determination without a scale. Few questions cause one to pause like walking into a pen of recently weighed cattle and being asked “What’s that steer weigh?”
As an opinionated, yet objective person, estimating weight without a scale causes internal tension, compelled to offer my guess yet knowing the true answer can be easily measured. One of my favorite books The Wisdom of Crowds by James Surowiecki recounts an example of how the averaged guess of an animal's weight by a crowd is more likely to correctly estimate an ox’s weight than any one individual. Reinforcing the power of multiple “measurements” diluting the errors.
A scale is only one of many ways weight can be determined. There are plenty of simple devices, measures and formulas one can use to determine weight. Some calf birth weights are determined using tapes to measure a foot or heart girth. Alternatively plenty of calves are “weighed” with one eye sizing up the calf while keeping the other on the cow.
These biometric measurements and the role technology can be used to enable their collection were the focus of Wang’s article. Several body measurements are correlated with weight, these often differ across species and may vary due to breed and stage of production. Some of the common measurements used previously to predict weight in cattle include body length, chest girth, heart girth / circumference, back width, distance flank to flank, hip height, hip width, rump length and width measurements, shoulder width / heights, as well as wither height.
This review highlighted how these measurements were originally manually obtained and transformed into predictive weights. While a tape around the heart girth is a useful alternative to a scale, measurements like these lack practicality.
The article outlined the evolution of these manual measurement techniques into a semi-automated system using photos and video cameras to automate the data capture. Despite the automation of data capture these methods remain difficult to expand due to the manual manipulation of the data required to separate animals from the environment and each other. Almost daily when proving “I’m not a robot” I realize sorting pictures of various objects is hard for computers.
Automated weight determination has real value, as a previous article we evaluated a system where animals were auto sorted after crossing a weighing station at a single feeding and water point. While this was an effective system to demonstrate the opportunity technology provides for automated weight capture and sorting, the singular location with limited portability provides practical challenges.
The review outlined the potential to use machine and deep learning computer models to elevate the video and sensor data in order to fully automate weight determination in extensive systems. We see examples of this machine learning technology advancing daily. Facial recognition started with humans and has already evolved to cows.
The University of Guelph research group highlighted the technical and practical challenges ahead. A key technical challenge outlined by the paper was adjusting weights for factors such as breed differences, performance potential and stage of production. Cloud-based connectivity enabling biometric sensors to sync with herd records and other “truth” sources to further inform the predictive models in real time will greatly reduce the development needed to overcome this technical limitation.
The challenge of technology acceptance by producers seems the simplest to address. Imagine capturing a check weight on a pen of yearlings by setting up a camera and sensor at the water tank. The sensors and algorithms go to work scanning the cattle, taking repeated measures and in a week you are ready to manage and market to an individual animal level. I suspect pen scales are going to look a lot different in the future.
