Automated oestrus detection method for group housed sows using acceleration measurements

A pilot experiment was carried out to define a method for oestrus detection for grouphoused sows based on sows' acceleration measurements. Acceleration data was measured for five sows, in three dimensions, four times per second during twenty days. The oestrus detection method is based on hourly averages of acceleration measurements, modeled using a univariate Dynamic Linear Model combined with a Kalman Filter. Alarms are produced when the observations deviate too widely from the model's estimated level. The same deviation patterns are observed for two sows out of three sows that came into oestrus. The data collection and the detection method are discussed and perspectives for improving the detection model while scaling up this pilot experiment are presented.