Animal Husbandry and Feed Science ›› 2025, Vol. 46 ›› Issue (2): 59-70.doi: 10.12160/j.issn.1672-5190.2025.02.008

• Animal Production and Management • Previous Articles     Next Articles

Automatic Recognition of Lying Postures and Analysis of Thoracoabdominal Movement Patterns in Parturient Dairy Goats

FENG Siyuan, AN Xiaoping, WANG Yuan, QI Jingwei   

  1. College of Animal Science, Inner Mongolia Agricultural University/Key Laboratory of Smart Animal Husbandry of Universities in Inner Mongolia Autonomous Region/Integrated Research Platform for Smart Animal Husbandry in Universities of Inner Mongolia Autonomous Region/Forage Technology Research Center for Herbivorous Livestock in Inner Mongolia Autonomous Region/Dairy Goat Breeding and Farming Technology Research Center of National Dairy Industry Technology Innovation Center, Hohhot 010018,China
  • Received:2025-01-10 Published:2025-07-09

Abstract: [Objective] To achieve automatic recognition of lying postures in parturient dairy goats using the You Only Look Once version 5s (YOLOv5s) model and analyze thoracoabdominal movement characteristics using the Farneback optical flow algorithm, thereby providing technical support for precise management of dairy goat parturition. [Methods] The YOLOv5s model was employed to classify lying and standing postures of parturient dairy goats, with model performance evaluated by precision (P), recall (R), and mean average precision (mAP). After video recognition, 20 Saanen dairy goats were divided into two groups based on parturition duration: Group A (parturition duration <30 min) and Group B (parturition duration ≥30 min). The Farneback optical flow algorithm was used to extract thoracoabdominal movement parameters (velocity, amplitude, duration of single movement, and frequency), and differences in movement patterns between the two groups were compared. [Results] ①The YOLOv5s model achieved P values of 98.4% and 98.3% for lying and standing posture recognition, respectively, with a false positive rate <2%, indicating minimal misjudgment risk; R values was 95.3% and 94.6%, with a missed detection rate <6%, demonstrating excellent detection coverage; mAP reached 96.3% and 95.2%, reflecting stable comprehensive performance and strong robustness. ②Optical flow analysis showed that the mean thoracoabdominal movement velocity in Group B was 5.358 px/s, significantly higher than that in Group A (2.461 px/s, P<0.05); the mean movement amplitude in Group B was 6.104 px, extremely significantly higher than that in Group A (2.280 px, P<0.01); the mean duration of single movements was 4.687 s in Group B and 4.272 s in Group A, with no significant difference (P=0.35); Group B showed a significantly higher movement frequency (45.67 times) compared to Group A (12.92 times, P<0.01), with reduced rhythmicity, indicating that parturition difficulty increases with prolonged parturition duration. [Conclusion] The synergistic application of the YOLOv5s model and Farneback optical flow algorithm enabled precise recognition of parturient dairy goat postures and accurate quantification of thoracoabdominal movements. This technology can be integrated into farm parturition early-warning systems to identify abnormal parturition behaviors in real time, reduce the risk of dystocia, and provide technical support for intelligent dairy goat management.

Key words: dairy goat, thoracoabdominal movement patterns, posture recognition, YOLOv5s model, Farneback optical flow algorithm

CLC Number: