Analysis · Technology
What a Cow Video Can Predict—and What the AI Test Leaves Open
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
A minute of video showing a cow’s face could contain clues about its milk production. In a 2021 study published in Sensors, Sigfredo Fuentes and colleagues at the University of Melbourne used camera measurements and a neural network to predict milk yield, composition and eye temperature. The revealing detail is how those predictions were tested.
The team collected 282 recordings from 102 Holstein-Friesian cows at one robotic dairy facility across four winter dates. Software estimated heart rate from brightness changes around the eyes and breathing rate from the nose region. Head movement measurements and weather information also entered the model. Thermal imaging supplied reference eye temperatures, while farm records supplied milk measurements.
Researchers allocated 150 recordings to model development, splitting these into 105 for training and 45 for testing. They evaluated the resulting model on the remaining 132 recordings. Across those additional recordings, predicted and observed values had a reported correlation of 0.93, pooled across five outputs: two milk yield measures, fat, protein and eye temperature.
That figure does not mean every prediction was 93% accurate. Pooling outcomes with different scales also leaves uncertainty about performance for each individual target. Another limitation concerns independence: cows contributed repeated recordings, and the methods describe random allocation of recordings without establishing that each cow appeared in only one group. Performance on unfamiliar animals therefore remains uncertain.
The setting further narrows the result. Measurements came from one farm in cool conditions, and cows showing high stress were excluded from recording. Heat stress monitoring and broader welfare applications were proposed uses, not demonstrated outcomes. The predictive associations do not establish that changing breathing, heart rate or movement would change milk production.
For deployment, this study offers a candidate monitoring method. Its practical value would depend on testing unfamiliar cows, farms and seasons, reporting errors separately for each output, and showing whether the information improves farm decisions. A strong combined correlation begins that evaluation; it does not establish operational benefits.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Biometric Physiological Responses from Dairy Cows Measured by Visible Remote Sensing Are Good Predictors of Milk Productivity and Quality through Artificial Intelligence Sigfredo Fuentes, Claudia Gonzalez Viejo, Eden Tongson …
