Parturition detection in sows as test case for measuring activity behaviour in farm animals by means of radar sensors

Abstract

General activity behaviour can be an indicator of numerous conditions in farm animals depending on the context of the measurement. Currently, most activity sensors established in livestock farming are optical or animal-attached sensors. Animal-attached sensors are hard to manage especially with smaller species. Optical sensors are susceptible to reduced sensitivity from dirt and to effects of changing light conditions. In scenarios where the spatial distribution of the activity is not relevant, simple stationary sensors which directly measure activity in an area may be beneficial as the signals are easy to interpret and hence require little computational power. The present work describes a proof of concept study for the utilisation of radar sensors for this purpose. In order to assess the effective sensitivity of the sensors under practical conditions, detecting the parturition-related activity increase in sows was chosen as benchmark. A simple classifier was implemented, detecting deviation in activity based on data from 15 sows, using day-to-day activity differences and the cumulative sum technique. This classifier achieved 92% sensitivity with 80% specificity and had a receiver operating characteristic with 95% area under the curve. This is comparable to the classification performance that has been achieved with animal-attached accelerometers and stationary light barriers. No negative effects from dust or other environmental conditions were observed. Additional studies will be necessary to verify this result and to show the suitability of radar sensors for additional application scenarios and other species.