Source: Beef Improvement Federation
“OneCup AI has designed an artificial intelligence (AI) technology using computer vision that is named BETSY, which stands for Bovine Expert Tracking and Surveillance,” explained Mokah Shmigelsky, OneCup AI. Shmigelsky was a featured speaker during the 2023 Beef Improvement Federation (BIF) Symposium July 5 in Calgary, Alberta, Canada.
According to Shmigelsky, BETSY was created because of the lack of permanent traceable identification in the livestock industry. Visual Identification was the first developed product followed by seven other value propositions.
“As we went through the developmental process, we discovered there was a lot of different things on farm that producers didn’t have access to because many of these processes are very manual,” Shmigelsky said.
In the beef and dairy cattle industry, the use cases that were identified were calving, feed tracking, shipping and tracking, and estrous and breeding. Shmigelsky said cattle tend to leave the herd or not face the camera making it difficult to collect data.
“In computer vision models you need to have a robust data set and all the data needs to be annotated as well,” Shmigelsky said. “Instead of tracking individual symptoms we were going to track what the animal’s behavior was.”
Once the animal is detected, BETSY identifies different behaviors and what is going on with the animal and alerts the producer. The user interface is an easy-to-understand system that gives the producer information graphical and visually along with alerts.
BETSY brain is broken down into levels that classify each animal by detection, bounding box, key points, identification, instance and time-series.
“Level 0 is detection it identifies different animals in the scene as well as humans and vehicles,” Shmigelsky said. “Level 1 is the bounding box it depicts the entire body of the animal and pulls out the relevant pixels needed for the next levels. Level 2 has 52 key points and angles which allow us to identify limping and changes in the animal. Level 3 is identification and can pair the final data to an individual animal as well as read the tag. Level 4 gets into the behavior and growth changes in the animals.
OneCup AI worked closely with the Canadian Angus Association to identify hooves, claws, udders and teats as the highest importance phenotypes. To identify these, they created additional points to the hooves and udders as well as created a 3-D visual. Clean animals made it easier to get a good ratio and analysis. Where mud and walking through grass effects accuracy.
“We are working on getting a more accurate reading so we can identify more than just the good things,” Shmigelsky explained.
Watch Shmigelsky’s full presentation: