Deep-learning-based behavioral time budgets for sows with high and low piglet mortality rates

S. S. Ferziger, I. C.F.S. Condotta, T. M. Brown-Brandl, Y. Shi, G. A. Rohrer

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Piglet crushing is a leading cause of pre-weaning mortality in piglets. Piglet crushing occurs when the sow lies down or rolls over. Although the use of farrowing crates can reduce pre-weaning piglet mortality, substantial pre-weaning mortality persists. Better understanding of the postures and behaviors of sows during the pre-weaning period is needed in order to develop better approaches to reducing pre-weaning mortality due to piglet crushing. The objectives of this study were to 1) create an object detection model using YOLO (You Only Look Once) to detect postures in sows, 2) Determine time budgets for postures of sows during the pre-weaning period using the pre-trained model, and 3) correlate time-budgets with pre-weaning mortality. Digital images of sows and piglets during the first four to seven days after farrowing were labeled for postures and behaviors. A YOLOv8 model was then trained using the labeled images. Time budgets of 4 postures for 6 sows during the first three days after farrowing were determined using the pre-trained YOLOv8 model. The model was capable of detecting four sow postures (kneeling, sitting, standing, and lying) with an overall mean average percent (mAP) of 0.979 at 0.5 intersection over union (IoU). The mAP for all classes was above 0.979 at 0.5 IoU and above 0.83 at 0.5-0.95 IoU. These results are sactifactory to automatically perform behavioral analysis. There were no statistical differences between high mortality and low mortality animals for time spent at each posture, but slightly more time spent kneeling and lying and less time spent sitting and standing was observed for low mortality animals.

Original languageEnglish (US)
Title of host publication2023 ASABE Annual International Meeting
PublisherAmerican Society of Agricultural and Biological Engineers
ISBN (Electronic)9781713885887
DOIs
StatePublished - 2023
Event2023 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2023 - Omaha, United States
Duration: Jul 9 2023Jul 12 2023

Publication series

Name2023 ASABE Annual International Meeting

Conference

Conference2023 American Society of Agricultural and Biological Engineers Annual International Meeting, ASABE 2023
Country/TerritoryUnited States
CityOmaha
Period7/9/237/12/23

Keywords

  • computer vision
  • posture recognition
  • Precision livestock farming

ASJC Scopus subject areas

  • Agronomy and Crop Science
  • Bioengineering

Fingerprint

Dive into the research topics of 'Deep-learning-based behavioral time budgets for sows with high and low piglet mortality rates'. Together they form a unique fingerprint.

Cite this