Smart Dairy Farming: A Mobile Application for Milk Yield Classification Tasks.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Dairy production relies on accurate milk yield assessment for management.
• Traditional methods can be labor-intensive and impractical in certain field conditions.
• Automated systems offer potential for efficient and accessible data collection.

Purpose of the Study:

• To develop and evaluate a lightweight, image-based deep learning model for classifying dairy cow milk yield.
• To automatically detect the udder region for yield category prediction.
• To create a practical tool for field-level assessment by non-specialist users.

Main Methods:

• Utilized a YOLOv11 architecture for object detection and classification.
• Trained the model on a public dataset of cow images with 305-day milk yield records.
• Established thresholds for low, medium, and high-yield classes and performed 30 training runs for robustness.

Main Results:

• The model achieved an overall precision of 0.408 ± 0.044, recall of 0.739 ± 0.095, and mAP@50 of 0.492 ± 0.031.
• The low-yield class showed the highest performance metrics.
• Misclassifications were primarily observed near class boundaries, highlighting the need for consistent image acquisition.

Conclusions:

• The study demonstrates the practical feasibility of using vision-based deep learning models in dairy production.
• The developed model and mobile application can aid decision-making, particularly in resource-limited settings.
• Automated udder detection offers a viable approach for non-specialist milk yield assessment.