Related Experiment Video
Updated: May 25, 2026
07:34
Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies
Published on: November 7, 2025
PickAMoo: LIDAR-enhanced mask R-CNN segmentation for precision weight estimation in dairy cattle using smartphone
Oleksiy Guzhva1, Emma Ternman2, Mikaela Lindberg3
1Department of Biosystems and Technology, Swedish University of Agricultural Sciences, Box 103, 230 53, Lomma, Sweden. oleksiy.guzhva@slu.se.
Scientific Reports
|May 23, 2026
Summary
A new smartphone app uses computer vision to estimate dairy cattle weight, improving farm management. This tool offers an accessible, low-cost alternative to traditional methods for monitoring animal health and productivity.
Area of Science:
- Agricultural Science
- Computer Vision
- Animal Science
Background:
- Accurate body weight and condition data are crucial for dairy farm management, including nutrient calculations, health monitoring, and breeding assessments.
- Traditional methods of weighing and assessing body condition in dairy cattle are labor-intensive and infrequent in practice.
- Existing automated solutions often rely on expensive multi-camera or 3D systems, limiting their farm-level applicability.
Purpose of the Study:
- To develop a practical, smartphone-centered workflow for estimating dairy cattle live weight.
- To create an accessible and cost-effective tool for on-farm use, reducing reliance on manual measurements.
- To streamline data collection for improved farm decision-making.
Main Methods:
- A two-step workflow was developed, beginning with a Mask R-CNN segmentation model trained on 567 annotated cow images (F1 score 0.98).
- Body weight was categorized using a Gaussian Mixture Model, followed by training a leak-safe pipeline (Extra Trees) on discretized weight data.
- A smartphone camera app was utilized for image collection, with evaluation employing cow-level grouped splitting and PyCaret for cross-checking.
Main Results:
- The segmentation model achieved a high F1 score of 0.98.
- The tuned Extra Trees model demonstrated a macro-F1 score of 0.936 (95% CI 0.860-0.983) with a 4.2% error rate on a holdout dataset.
- The system was validated on 1080 images collected via the developed camera app, separate from the segmentation model training data.
Conclusions:
- The developed smartphone-centered workflow provides a practical and accurate method for estimating dairy cattle live weight.
- This approach offers a cost-effective and accessible alternative to traditional and complex automated systems for on-farm applications.
- The potential exists to further optimize the algorithm for a scalable, open-source smartphone application to support farm management.