Current Progressive Dairy digital edition

Digital data drives future of on-farm decision-making

Progressive Dairy Editor Karen Lee Published on 23 August 2019
Security camera

As we approach 2020, the ability to extract data from a dairy barn continues to grow. Advances in artificial intelligence are already being implemented on farms and can impact how farm management decisions are made.

“It’s more about farming data,” Robin Johnston said at the 2019 PDO Triennial Symposium, March 4-6, in Toronto, Ontario.



“What is a digital dairy?” Johnston asked. “In its truest sense, it’s a farm that’s information-enabled. It’s about searchable, queryable data in real time. That means datasets on-farm that can actually be useful while you are out in the field.”

With a background in computer science, Johnston is the chief scientific officer and co-founder of Cainthus. His work focuses on machine learning, a subset of artificial intelligence, where he creates algorithms and software that, when given a dataset, can teach itself.

“Machine learning learns for itself and recognizes patterns by repetition,” he explained.

Johnston’s work is in visual analysis, primarily through security cameras mounted in various locations in the barn.

They can look at bedding packs to know when they need to be changed. It can collect body condition scores and monitor how a cow drinks.


They have done a lot of work on feed to measure feed levels in the barn, see how well the ration is mixed, how it has been delivered, when push-ups should happen, where cows are eating and how much they are eating.

The system is able to recognize cows from all angles to understand movements.

All data collected is sent to a computer on-site for processing, then it gets assembled in the cloud. It is available in real time and accessible from a phone.

“If a cow gets knocked down in a pen, your phone will go off in about a minute,” Johnston said.

Infrared lights on the cameras illuminate the cows at night, so data collection continues even when people are not in the barn.

How does it work?

There are many lanes in how artificial intelligence operates, but Johnston only highlighted the four main avenues applicable here:


  1. Training – Existing datasets on the farm, pictures and data the cameras collect can be merged into one pool to teach the system new things.

  2. Detection – Finding cows in the barn and understanding who’s who

  3. Logic – This is what the system knows. “This is where it gets really interesting,” he said. “Let’s say somebody comes up with an idea that when the left ear twitches, that’s 40 [quarts] of milk. All you need is to see that, and we can reteach the system what that means. From an office, we can push that new intelligence to the farm and, all of a sudden, that system on the farm is smarter. So it’s constantly evolving.”

  4. Data – The point is to collect data that is useful and actionable in real time.

“One of the beauties of using artificial intelligence systems is: They are adaptable,” Johnston said. The cameras can be positioned in many different places. The system can also be altered at any time. What is in there now doesn’t necessarily have to be there a year from now. This is different than wearable data collection devices that typically have the same task through the life of the unit.

On-farm troubleshooting

“All of this comes back to efficiency and problem-solving on the farm,” he said.

In one instance, a camera was positioned over a water trough that had a sorting gate across it. When the gate was closed, the cows chose to drink from the other trough, but it didn’t have the capacity to support them. There was a large portion of the population that wasn’t drinking for eight hours of the day. When the gate opened up, the cows rushed to the trough because they were thirsty, and a pattern developed.

“You can see why production in the barn wasn’t as good as it could have been. This is just a simple thing you can catch with systems like this,” he said.

In another scenario, the person feeding the cows put in too much of one ingredient, which caused the mixer to not mix properly and was therefore unable to deliver feed consistently. The system was able to see the hot spots in the middle of the pen where most of the cows were eating. Some cows were left to pick what they could from the outside areas with little food and improper nutrients.

After one producer had regrooved concrete in part of the pen, the system showed how cows preferred to eat there, and he decided to regroove the entire pen.

On a different farm, a security light comes on at a certain time of night. When it does, cows come to eat. When that light goes off, cows go back to their beds. “Proper lighting in the barn is not a joke; it does affect production, and we can show it,” Johnston said.

Similarly, consistent feed delivery matters. In another scenario, the cows were fed 90 minutes later than usual. They still hit a peak intake at the same time of day, but there was a large difference compared to the height of the peak the previous day. Milk production dropped and was down the subsequent day as well.

In his last example, a particular farm was notified every day how the cows were eating, and Johnston was included on these notifications. “I noticed that intake time was going down, and the cows weren’t eating as much,” he said. After letting it slide for a little while, he went to the farm and talked with the nutritionist, who said nothing had changed. Johnston returned to the farm the next week, where he learned the producer had changed protocols on how he was defacing the haylage bunker. By week three, the ration was changed to compensate for the problem.

“The point of all of this was: The producer’s milk production was going down a lot, but he couldn’t pinpoint the problem. The nutrition team was doing a great job, but nobody told them there were any problems going on. We caught it within days and could help everyone fix the issue,” he said.

“These are the types of things that are going to help everyone make more money by being more efficient,” Johnston added.

This farm was losing 7 pounds of milk in that pen for almost three weeks. This type of system can help producers quantify the effects of a protocol change on the farm.

“There is a cause and effect of everything you do on your farm. You need to be able to quantify what you’ve done. If you can’t measure it, you can’t fix it,” he said.

Future possibilities

Johnston is working on a university trial to detect cows in heat based on how they eat. “We can catch them 24 hours before a typical wearable can,” he said. “Each cow has its own dataset, so we can capture subtle changes.”

They are teaching the system to identify sorting of feed to detect mixing and ration problems. One producer asked them to identify cows that were eating aggressively.

Eartag reading has the potential to replace RFID in parlors.

Remote access for veterinarians will help them check on a post-surgical cow to see if it is eating, drinking and lying down without returning to the farm.

Integrating visual data with other information systems will help find causative links and streamline systems on the farm.

“There are certain vendors in the world that are waking up to the fact we can open and close gates by what the cows look like. Very shortly, the barn will be reacting to the cows as opposed to putting more and more stuff on the cows to make gates and robots and everything else open and close,” Johnston said.

Artificial intelligence systems will continue to find their way onto farms because cameras can see things sooner and more consistently than humans. “Cameras don’t get tired. They don’t take days off. They see 24 hours a day,” he said.

When they can be integrated with on-farm data, new systems can be created. “It is constantly evolving and changing. That’s the whole point of this – is the evolution of datasets,” he said.

These systems aren’t meant to replace people but, instead, augment their ability to do their job. Johnston explained, “We look at cows and see them all the time. [These systems] catch the little things that, otherwise, people miss.”

It does mean identifying problems and areas where management isn’t as good as one might think. “For as much as a good job as you’re doing, there’s oftentimes sobering truths about how you are working on the farm,” he said.

“That’s why this is important,” Johnston added. “We need to make data-driven decision-making. We need to separate noise from data, which is what artificial intelligence does.”

New data collection systems can measure more on the farm today than what was achieved before. When it is actionable and in real time, it can lead to faster decision-making and more progress.  end mark

PHOTO: Security cameras capture visual images used by artificial intelligence software to monitor and measure what is happening in the barn. Courtesy photo.

Karen Lee
  • Karen Lee

  • Editor
  • Progressive Dairy
  • Email Karen Lee