Edge-AI Enabled Acoustic Monitoring and Spatial Localisation for Sow Oestrus Detection.

Area of Science:

• Precision Livestock Farming
• Artificial Intelligence in Agriculture
• Animal Reproductive Physiology

Background:

• Accurate sow oestrus detection is critical for optimizing reproductive efficiency and minimizing non-productive days in swine operations.
• Traditional methods like manual observation are labor-intensive and subjective.
• Existing cloud-based deep learning solutions present challenges in latency and data privacy within intensive farming environments.

Purpose of the Study:

• To develop an edge-intelligent monitoring system for real-time sow oestrus detection.
• To integrate deep temporal modeling with sound source localization for precise monitoring.
• To address the limitations of manual observation and cloud-based AI in pig farming.

Main Methods:

• A lightweight Stacked-LSTM model was selected and deployed on ESP32-S3 hardware using a hierarchical screening strategy.
• The model was trained and calibrated with an acoustic dataset, validated against serum reproductive hormones (FSH, LH, P4).
• A sound source localization algorithm (GCC-PHAT) was integrated for mapping vocalizations to individual gestation stalls.

Main Results:

• The optimized model achieved 96.17% classification accuracy with a low inference latency of 41 ms.
• The system successfully mapped vocalization events to specific gestation stalls.
• The "edge recognition-cloud synchronization" architecture demonstrated robustness and cost-effectiveness in laboratory tests.

Conclusions:

• The developed edge AI system provides a reliable and efficient solution for real-time sow oestrus detection.
• This technology offers a robust technical framework for precision management in smart livestock farming.
• The system overcomes latency and privacy concerns associated with cloud-based solutions.