Automated Detection of Feed Spilling Behavior for Rabbit Abortion Monitoring

Automated monitoring of breeding activities in intensive rabbit farming is crucial for timely intervention against events affecting production efficiency, such as feed spilling caused by abortion. Traditional manual inspections are inefficient and often fail to detect subtle abortion signs such as bloodstains or expelled fetuses, as well as small, camouflaged feed pellets scattered across complex manure boards. This study proposes a multi-task detection approach based on the YOLOv8 architecture to address these challenges. We built a custom dataset of 4,985 high-resolution RGB images annotated for bloodstain, fetus, and feed pellets, and trained a multi-task pipeline for detection tasks, including the YOLOv8n for detecting abortion signs, and YOLOv8n-P2 for feed pellets. The multi-task detection results are fused via conditional filtering to localize feed pellets caused by abortion. Our solution achieves accurate detection results with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5$</tex> of 0.81, precision of 0.87, and recall of 0.79.