Automated Analysis of Dairy Cow Drinking Behaviour Using Computer Vision: Developing and Integrating Cow Detection and Drinking Behaviour Classification Components
Summary
Monitoring the drinking behaviour of dairy cows provides valuable insights into their health and welfare. However, establishing the relationship
between water intake and factors like milk production has been challenging
due to limitations in data collection and the amount of research on this subject. Computer vision offers a promising solution for automated monitoring
of cow drinking behaviour.
A system for cow detection and drinking behaviour classification using deep learning techniques was presented. A YOLOv10 model achieved
99.1% AP-50 and 87.1% AP50-95 for cow detection, while an EfficientNetV2-
S model attained 88.6% accuracy for binary classification of drinking behaviour. When tested on a 1-hour video, the system measured drinking
time with 92.5% precision and 92.0% recall, demonstrating its effectiveness
for automated analysis.
Integration of this system with cow identification components will enable monitoring of individual free water intake, providing valuable data for
studying the relationship between free water intake and milk production.
The rigorous testing and evaluation conducted in this work paves the way
for practical application in precision livestock farming.
Future research should explore the addition of spatial-temporal components to further improve performance and investigate the impact of different camera viewpoints. Ultimately, this system contributes to the advancement of animal welfare and farm management by enabling detailed analysis
of drinking behaviour.