Assessing computer vision as a tool to automate dairy cattle social behavior analysis

As the world population continues to increase, the need for animal-derived products also continues to grow. Precision livestock farming (PLF) technologies have been able to help with these increasing demands on dairy farms. Computer vision is one of these PLF technologies that has been used to collect health, behavior, environment, and welfare information. This technological development is essential because welfare standards within the dairy industry continue to rise, thus causing a larger labor force to be required to meet care standards. Automated social behavior classification can be a way to combat these issues and lead to different farm management strategies incorporating animal needs on a more individualized level. The goal of this thesis was to review how current methods of dairy cow behavior are collected, create an automated behavior classification model based on computer vision to identify both the active and passive cows during social interactions, as well as analyze cow behavioral patterns in terms of social network and environmental factors. A Yolo v8 model was successfully developed to classify both the active animal and passive animal/object for five behaviors which included body butting, head butting, licking, rubbing stall, and throwing feed. The F1 scores for each behavior follow: body butting at 69.79%, head butting at 66.67%, licking at 76.79%, rubbing stall at 84.21%, throwing feed at 100.00%. It was determined that environmental factors and pregnancy did have an effect on the five observed behaviors. Body butting initiation behaviors were impacted by the average hourly temperature during the observation period (P = 0.0402). Head butting initiation and frequency were affected by pregnancy status of the cow (P = 0.001863). Licking initiation and frequency were affected by both average hourly temperature during the observation period (P = 0.01138) and daylength (P = 0.01338). Licking behavior initiation was affected by pregnancy status of the cow (P = 0.001666). Rubbing stall initiation and frequency behaviors were strongly affected by daylength (P = 1.156e-8). Throwing feed initiation behaviors were impacted by the average hourly temperature during the observation period (P = 0.04262) and the daily temperature high (P = 0.02133). All in all, each behavior observed was impacted by at least one of the tested variables which concluded that many environmental factors do impact behavior.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Animal Sciences

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Bone, Breanna J.

Contributors dc:contributor

• Condotta, Isabelle C.F.S.
• Cardoso, Felipe
• Green-Miller, Angela

• Dairy Cows
• Social Interactions
• Computer Vision
• Machine Learning
• Precision Technology

Statement dc:rights

• Copyright 2023 Breanna Bone

Language dc:language

ar, eng

Handle dc:identifier

https://hdl.handle.net/2142/122176

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F122176