Video-Based Feeding Demand Sensing and Offline Feeding Strategy Evaluation in Group-Housed Pigs: A Single-Pen...

Area of Science:

• Animal Science
• Agricultural Engineering
• Computer Vision

Background:

• Accurate sensing of group feeding demand is crucial for optimizing feeding strategies in large-scale pig farming.
• Current methods struggle with dense housing, leading to inaccurate visual recognition and reliance on suboptimal fixed feeding schedules.
• There's a need for real-time, reliable methods to assess group feeding behavior and align it with demand.

Purpose of the Study:

• To develop and evaluate a video-based workflow for sensing group feeding activity in pigs.
• To assess the feasibility of using video-derived data for evaluating feeding schedule allocation rules.
• To determine if behavioral signals can serve as a proxy for group feeding demand.

Main Methods:

• A workflow integrating rotated object detection, visual feeding intensity calculation, and video-derived feeding demand was developed.
• Synchronized video and feeder data were used to train and validate the detection models.
• Feeding strategies were evaluated offline using the video-derived demand data.

Main Results:

• Rotated bounding-box detection achieved high accuracy (mAP50-95 of 0.908) for feeding recognition in dense pen environments.
• Visual feeding intensity showed strong correlation (r = 0.80) with actual feed intake.
• The video-perceived hybrid strategy improved synchronization (0.611) and reduced mismatch rate (5.13%) compared to traditional methods.

Conclusions:

• Video-based behavioral signals can effectively serve as a group-level proxy for feeding demand in pigs.
• This approach supports the offline evaluation and optimization of behavior-aware feeding schedules.
• Further validation across diverse conditions is needed for practical deployment.