Publication
Authors
Anita Z. Chang, Senior Postdoctoral Research Fellow, Institute for Future Farming Systems, School of Health, Medical, and Applied Sciences, CQUniversity Australia (CQU)
Eloise S. Fogarty, Senior Research Officer, CQU
Luis E. Moraes, former Assistant Professor, Department of Animal Sciences (ANSCI)
Alvaro García-Guerra, Assistant Professor, ANSCI
David L. Swain, Adjunct Professor, CQU
Mark G. Trotter, Professor, CQU
Quick Take
Animals can’t tell us when they don’t feel well, but they can show us. Behavior changes are often important indicators of animals’ health and well-being. In ruminants, or animals that digest their food by regurgitating and rechewing it, such as cattle and sheep, changes in rumination patterns can be signs of disease, females being in heat, and calving. Keeping track of these changes is important for farmers, but it’s almost physically and economically impossible to monitor livestock in current farm situations. Automating this process using on-animal sensors and machine learning is becoming increasingly popular, especially for large farming operations. Data from the sensors can be used to identify specific behaviors based on the animals’ movement.
A team of researchers from the College of Food, Agricultural, and Environmental Sciences and CQUniversity in Australia sought to find the best methods for accurately detecting rumination, especially for each individual animal as compared to the more common generic models. The researchers used ear tags on Angus crossbred cows that contained accelerometers to detect movements associated with rumination (very slight motions of the jaw muscles when cows chew, swallow, and regurgitate food). Similar to the way Fitbit-type sensors track a person’s steps, this technology allowed the research team to monitor each cows’ chews.
The team compared four machine learning methods: classification and regression tree (CART), random forests (RF), linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA). They also used six different fixed time intervals (epochs) and a mixed epoch that combined multiple fixed time epochs. Their use of ear tags may prove to be an accurate and cost-effective way of detecting behaviors in cattle and other livestock, which will likely improve our ability to maintain animal welfare and prevent diseases from spreading.
Results
-
Researchers concluded that rumination behavior can be detected with relatively high accuracy using an ear-tag accelerometer. The placement of the sensor tag may impact behavior detection, but more research is needed to investigate this.
-
This is the first paper to examine the difference between generic and individual animal models for rumination detection. It found that accuracy was significantly higher using an individual model compared to any generic model.
-
Using a mixed epoch and a classification and regression tree (CART) was the machine-learning method with the highest average accuracy as a generic model.
-
Mixed epochs had the highest accuracy and sensitivity overall, though the requirements for processing power, and, therefore, computational requirements, were higher. Mixed epochs may also be useful for detecting other complex behaviors and reducing the risk of false positives, but more research is needed on this.
-
The CART model took the shortest time to compute, only a few minutes or less regardless of time intervals, whereas the random forests (RF) model took the longest of the machine learning algorithms tested at nearly 45 minutes when using mixed epochs. Longer fixed time intervals (epochs) took less time to compute overall than many short epochs or mixed epochs.
-
More research using other species and conditions is needed to get a better idea of how well different machine-learning methods and models may work for various situations.
-
Commercial smart-tag retailers have been reviewing this research to understand how they can adapt it for use in their systems, which are currently being sold to farmers across the globe.
Citation
Chang, A. Z., Fogarty, E. S., Moraes, L. E., García-Guerra, A., Swain, D. L., Trotter, M. G. “Detection of rumination in cattle using an accelerometer ear-tag: A comparison of analytical methods and individual animal and generic models.” Computers and Electronics in Agriculture, 192, January 2022, 106595, https://doi.org/10.1016/j.compag.2021.106595