Answers are grounded in this paper and cite their sources.
In modern group-housed pig production, mounting behaviour provides observable evidence for prioritising reproductive observation, but its practical use is constrained by complex social interactions, frequent occlusion, posture variation, and the difficulty of transforming detected events into interpretable management information. This study proposes Pig-RDF, a three-layer smart livestock framework comprising perception, analysis, and report generation. The perception layer employs EBF-YOLO, a lightweight detector incorporating spatial-frequency modules, which achieved 99.1% precision, 98.5% recall, 99.2% mAP@0.5, and an inference speed of 109.6 frames per second on the internal test set. The analysis layer aggregates detections at the pen level and quantifies diurnal rhythm, multi-day patterns, and duration distribution to characterise mounting-behaviour evidence and prioritise follow-up oestrus observation. The report-generation layer integrates these descriptors into a large language model (LLM)-based assistant using GPT-4o, prompt engineering, and a predefined stepwise reasoning workflow. Human evaluation of 30 generated reports yielded an overall score of 4.80 ± 0.15 on a five-point scale, with no critical errors identified across the predefined checks. Overall, Pig-RDF establishes a data-traceable "video-in, report-out" workflow and demonstrates the feasibility of integrating deep learning-based sensing, behavioural pattern quantification, and LLM-assisted reporting for preliminary pen-level reproductive observation support based on mounting-behaviour evidence.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: