On-animal motion sensing using accelerometers as a tool for monitoring sheep behaviour and health status

Abstract

An opportunity exists to infer the physiological and physical state of an animal from changes in their behaviour. As resting, eating, walking and ruminating are the predominant daily activities of ruminant animals, monitoring these behaviours could provide valuable information for monitoring individual animal health and welfare status. Conventional animal monitoring methods have relied on visual observations of animals by human labour. This can only provide information on an animal's behaviour for the period in which they are being observed. Historically, observations could be made for long periods where shepherds were employed to observe their flocks nearly constantly. This is obviously no-longer feasible in the current livestock industry. Recently, with the advent of small, low power accelerometer technology, the ability to remotely monitor animal movement continuously has arisen. This is achieved through the application of on-animal inertia monitoring unit (IMU) sensors. This movement data might potentially lead to continuous behavioural monitoring of livestock. These devices have been developed for higher value livestock such as dairy cattle but little research or development has been directed towards their use in sheep. Previous work has evaluated collar and leg deployments however the sheep industry demands these devices be in an eartag form factor to align with current industry practices. Therefore, this thesis aims to evaluate the potential for using ear-borne accelerometer devices to detect and categorise key behaviours expressed by sheep. Deviation from normal patterns of behaviour may be used as an indicator of changes in individual health status. If behaviour can be categorised using the data collected by these body worn devices and radio telemetry incorporated, animal health could be monitored in near real time allowing early treatment intervention when necessary, ultimately improving on-farm productivity. Scoping work in this thesis identified the difference in acceleration signals between the basic sheep behaviours: grazing, walking and resting, giving potential for discrimination between behaviours with classification algorithms. Subsequently a successful behaviour classification algorithm was developed based on accelerometer data obtained from the ear deployment, yielding activity predictions similar to those obtained through visual observation. To apply this technology to a commercial application, a simulated lameness experiment was designed, where lame walking behaviour was discriminated from sound walking events successfully using the ear and leg modes of deployment. The final experiment investigated the application of ear deployed accelerometer devices to detect behavioural changes associated with increased infection by internal parasites, a disease of extreme economic importance within Australia. Animals with a higher faecal worm egg count were shown to have a lower probability of engaging in longer periods of activity, however this experiment was limited by a very mild level of infection. Overall this thesis demonstrates that sheep behaviour can be classified using an ear-mounted tri-axial accelerometer sensor, the first of its kind to date. It also explored the suitability of using time-series behavioural classification data as an early indicator of health and welfare issues. This work aims to link a previous "research only" technology in sheep, to a commercial application, a stepping stone towards bridging the gap between research and industry adoption.