Estimating vegetation index for outdoor free-range pig production using YOLO

Area of Science:

• Agricultural Technology
• Remote Sensing
• Computer Vision

Background:

• Outdoor free-range pig production requires efficient methods to monitor grazing impact.
• Assessing cover crop damage from free-ranging livestock is crucial for sustainable agriculture.
• Existing machine learning applications in agriculture primarily focus on pest and fruit detection.

Purpose of the Study:

• To quantitatively estimate grazing area damage in free-range pig production.
• To evaluate the effectiveness of Unmanned Aerial Vehicles (UAV) and deep learning for monitoring crop occupancy.
• To determine optimal grazing durations for free-range sows to prevent overgrazing.

Main Methods:

• Utilized UAVs with RGB sensors to capture aerial imagery of a 100x50m cornfield over two weeks.
• Processed images using a You Only Look Once (YOLO) deep learning model for corn condition detection.
• Employed extensive data augmentation techniques, generating over 24,000 training datasets from raw images.

Main Results:

• The YOLO model efficiently estimated corn occupancy rates in the grazing area.
• Significant corn disappearance was observed within nine days of sows grazing.
• Calculated that 20 sows in a 50x100m field (250m²/sow) require rotation after approximately five days.

Conclusions:

• UAVs combined with YOLO provide an effective method for assessing grazing damage in agricultural settings.
• Data augmentation is essential for training deep learning models when expert-collected data is limited.
• Findings provide practical recommendations for managing grazing intensity to ensure cover crop sustainability.