A Combined Offline and Online Algorithm for Real-Time and Long-Term Classification of Sheep Behaviour: Novel Approach...

Area of Science:

• Agricultural Engineering
• Animal Science
• Machine Learning

Background:

• Precision livestock farming relies on real-time monitoring for animal welfare and productivity.
• Long-term monitoring systems face challenges like concept drift, where system performance degrades due to changing conditions or data discrepancies.

Purpose of the Study:

• To develop and evaluate a novel algorithm that effectively handles concept drift in animal behavior monitoring.
• To improve the accuracy and reliability of real-time sheep behavior classification under dynamic environmental conditions.

Main Methods:

• A combined offline and online learning algorithm was developed to manage concept drift.
• Sheep behaviors were classified using data from an embedded edge device with tri-axial accelerometer and gyroscope sensors.
• The algorithm was tested for real-time classification of three key sheep behaviors.

Main Results:

• The proposed algorithm demonstrated improved real-time classification of sheep behaviors.
• The system effectively addressed concept drift, maintaining performance under changing conditions.
• This represents a novel application of such algorithms in precision livestock behavior monitoring.

Conclusions:

• The developed algorithm is a valuable mechanism for long-term, in-the-field monitoring systems in precision livestock farming.
• The approach enhances the ability to monitor animal welfare and productivity by accurately classifying behaviors dynamically.
• This study offers a significant advancement for real-time behavioral analysis in sheep farming.