Machine Learning for Dairy Cow Behaviour Classification

Type

Student thesis: Doctoral ThesisDoctor of Philosophy by Published Works

Original languageEnglish
Awarding Institution
Supervisors/Advisors
Award date23 Apr 2020
Links
Show download statistics
View graph of relations

Abstract

This thesis describes the use of machine learning (ML) techniques applied to data gathered from GPS receivers attached to pasture-based dairy cows for the purpose of automatic behaviour identification. Automatically identifying the behaviour of cattle will allow livestock practitioners to make more informed decisions on their management. Furthermore, daily behaviour data can be utilised for earlier disease diagnosis. For example, if the feeding duration of a cow is below its expected target then managers can intervene. Individual animal data were previously unattainable, with cattle usually managed on a herd basis. This thesis begins with an introduction that summarises the ongoing research in the field of precision livestock farming (PLF) and how farmers are implementing some PLF systems for the management of livestock. The main PLF systems discussed are those that incorporate on-animal sensors for the detection and classification of key behaviours associated with production and health. The main body of the thesis is divided into three experimental chapters. Chapter 1 (published in the Journal of Dairy Science) describes the development of a behavioural model of pasture-based Holstein dairy cows using data collected from GPS receivers and processed using ML techniques. Chapter 2 (published in Computers and Electronics in Agriculture) discusses a further modification to the behavioural model which improves its ability to categorise behaviours. Finally, Chapter 3 describes the use of a data partitioning technique often used for timeseries analysis as an alternative method for the development of behaviour prediction models of dairy cows. Chapter 3 was published in the journal Biosystems Engineering. The thesis concludes with a discussion of each chapter in light of the wider research and highlights some necessary areas for further work.

Keywords

  • PLF, machine learning, cattle behaviour, movement ecology, livestock disease