Validation of accelerometers to automatically record sow postures and stepping behaviour

Two studies were performed to develop and validate an automated method of detecting postures and stepping behaviour in sows. In the first study, two accelerometers were simultaneously tested on 23 multiparous sows to detect the following postures: standing, sitting, lying ventrally and lying laterally. First off, a data set from 11 sows was used to establish the methodology and algorithm to automatically detect postures, and a second set from 12 sows was used for validation purposes. Sows were housed in gestating stalls, pens, farrowing crates or farrowing pens with straw. One accelerometer was fastened to a hind leg and the other to the back of the sow (between shoulder blades). The data loggers recorded the acceleration on three axes every 5 s for 6 h; these data were then converted into degrees of tilt, which were used to discriminate between postures according to angles determined with the first data set. Based on video observations, sows spent an average time of 23.1 ± 0.1% standing, 24.6 ± 0.2% lying ventrally, 48.1 ± 0.3% lying laterally and 4.2 ± 0.1% sitting. Sensitivity values (the extent to which the accelerometers correctly detect each posture) for standing, lying ventrally, lying laterally and sitting were 100 ± 0.01%, 94 ± 0.04%, 91 ± 0.2% and 50 ± 0.4%, respectively. Specificity values (the extent to which the accelerometers correctly identify true negatives) were above 90% for all postures. The results suggest that individual calibration would have increased the performance of the accelerometers for sitting. The second study was performed to validate the use of accelerometers for counting hind limb stepping behaviour around feeding in 10 sows. Animals were housed either in gestating stalls or in pens and had an accelerometer fastened to one rear leg. The data logger recorded the acceleration on the vertical axis 10 times per s for 30 min, starting at the time of feeding. The accelerometer data was compared to video observations and 1448 steps were assessed in total. The average sensitivity value was 95 ± 0.04% with an error of 5 ± 0.03%. In conclusion, accelerometers can be successfully used to detect postures and the number of hind limb steps in sows.