Automatic Position Detection and Posture Recognition of Grouped Pigs Based on Deep Learning

The accurate and rapid detection of objects in videos facilitates the identification abnormal behaviors pigs introduction preventive measures to reduce morbidity. In addition, effective pig algorithms provide a basis for behavior analysis management decision-making. Monitoring posture can enable precursors diseases timely manner identify factors that impact pigs’ health, which helps evaluate their health status comfort. Excessive sitting represents when are frustrated restricted environment. present study focuses on automatic recognition standing lying grouped pigs, shows lack posture. main contributions this paper as follows: A human-annotated dataset standing, lying, postures captured by 2D cameras during day night barn was established, simplified copy, paste, label smoothing strategy applied solve problem class imbalance caused among dataset. improved YOLOX has an average precision with intersection over union threshold 0.5 (AP0.5) 99.5% 0.5–0.95 (AP0.5–0.95) 91% position detection; AP0.5 90.9% AP0.5–0.95 82.8% recognition; mean (mAP0.5) 95.7% (mAP0.5–0.95) 87.2% all recognition. method proposed our improve effectively, especially recognition, meet needs practical application farms.