Back to search

Michigan State University

An automated 3D pose estimation system for sow health monitoring

Abstract

dc:description

Pork ranks as one of the most consumed meats globally, presenting both a challenge and an opportunity to both improve the care of swineherds and to increase efficiency of production. Conditions such as lameness and poor body composition impair productivity and animal welfare, while current assessment methods are subjective and labor-intensive, resulting in slower production and ambiguous quality classifications. Precision Livestock Farming (PLF) proposes using technology to monitor animals, assess health, and apply data-driven interventions to increase welfare and produce higher quality products. For sows, body shape and motion characteristics provide important health indicators that could be assessed automatically through appropriate PLF sensors and techniques.This thesis developed a sow PLF health assessment device using artificial intelligence, including both a hardware system and a novel training algorithm. First, a data collection device, the SIMKit, was built using modern dense depth sensors, which can reveal detailed shape characteristics of a sow. Second, a new annotation method called, Transfer Labeling was developed, enabling the semi-automated annotation of a large dataset of sow depth images. This dataset was used to train a convolutional neural network (CNN) to detect and track pig poses. Results show that Transfer Labeling produces annotations with sub-centimeter accuracy with much-reduced human effort. It is anticipated that this will lead to much-improved sow health monitoring.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Yik, Steven
Contributors dc:contributor
  • Morris, Daniel
  • Benjamin, Madonna
  • Lavagnino, Michael
  • Srivastava, Vaibhav

Subjects

dc:subject × 6

Rights

dc:rights
Statement dc:rights
  • In Copyright
Language dc:language
English

Identifiers

dc:identifier.*
Identifier
etd:48526
isbn:9798645450328
oclc:1164816715
umi:27961474
local:Yik_grad.msu_0128N_17553

Chain of custody

source
Harvested from
Michigan State University
Base URL
d.lib.msu.edu/oai
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Yik, Steven. An automated 3D pose estimation system for sow health monitoring. 2020. https://doi.org/doi:10.25335/np5a-0f82