LETI: Neural Network Tracks Deviations in Cow Behavior
The number of cows on modern dairy farms can be in the thousands. In such conditions, early tracking of diseases or ailments in animals at an early stage becomes an important factor in reducing the economic costs of equipment and veterinary services.
Earlier, LETI scientists proposed to monitor the condition of cows using a neural network model that recognizes the main characteristics of their behavior based on video recordings. For this purpose, in 2020, several dozens of cows were filmed on a dairy farm of EkoNivaAgro in the Voronezh Region. Using the collected data, the neural network model learned to determine what cows look like and how they differ from each other, and this prototype won a grant from the LETI Youth Innovation Projects Contest.
“Now we’ve added an analysis of feeding behavior. That is, now the neural network not only understands that there is a particular cow in front of it but also how often and how much it eats. And the program can register even minimal deviations from normal behavior and report them to the operator – this is important for identifying ailments at an early stage.”
Eugene Shalugin, a first-year master’s student of the Faculty of Computer Science and Technology
Today, there are few databases of pet videos that are publicly available to researchers, so the most important task for the researchers was to process data on cow behavior in a way that would make it suitable for training a neural network. Then the researchers worked with those video fragments in which the animals ate, tracing the patterns of trajectories of body parts movements. In addition, they purchased three special IP cameras to capture new footage using the grant won in 2020.
“Analysis of behavior based on video from three cameras at once allows us to more accurately record the trajectories of living objects in the coordinate system. After processing the data set, our neural network can be expanded to monitor not only feeding behavior but also the way each individual cow walks, stands, and lies. This extends the possibilities for diagnosing the condition of animals.”
Dmitry Kaplun, Associate Professor at the Department of Automation and Control Processes
Scientists plan to go to a dairy farm soon to collect more information. Also, to assess the correctness of the proposed solution, it needs to be tested on video data with cows from other farms. The ultimate goal is to create a universal software that can detect abnormal behavioral patterns in humans and different types of pets.