Toward the automatic detection of social interactions in gestating sows using image analysis data

Area of Science:

• Animal behavior
• Ethology
• Precision livestock farming

Background:

• Monitoring social interactions in group-housed gestating sows is vital for welfare assessment.
• Manual observation of sow interactions is resource-intensive and time-consuming.
• Automated systems are needed to efficiently analyze sow social behavior.

Purpose of the Study:

• To assess the feasibility of using image analysis software to automatically detect and classify social interactions in gestating sows.
• To determine if sow postures and key point data can replace manual observation for welfare monitoring.

Main Methods:

• Video recording of 39 gestating sows across two pens over two gestations.
• Manual annotation of positive, negative, and non-interaction events using software.
• Analysis of sow postures, key point coordinates, distances, and movement speeds.
• Application of decision trees to identify key variables for interaction detection and classification.

Main Results:

• Automated detection of interaction start/end achieved 88% accuracy using distances and postures.
• Interaction valence (positive/negative) was classified with 74-80% accuracy based on movement.
• Specific interaction orientations, like nose-tail, were detected with high accuracy (94% F-score).

Conclusions:

• Image analysis software shows significant potential for automated detection and classification of social interactions in gestating sows.
• This technology can reduce the need for manual observation, offering a more efficient welfare monitoring solution.
• Further development of the software can enhance precision livestock farming practices.