Improved cattle behaviour monitoring by combining Ultra-Wideband location and accelerometer data.

Cattle behaviour is fundamentally linked to the cows' health, (re)production, and welfare. The aim of this study was to present an efficient method to incorporate Ultra-Wideband (UWB) indoor location and accelerometer data for improved cattle behaviour monitoring systems. In total, 30 dairy cows were fitted with UWB Pozyx wearable tracking tags (Pozyx, Ghent, Belgium) on the upper (dorsal) side of the cow's neck. In addition to the location data, the Pozyx tag reports accelerometer data as well. The combination of both sensor data was performed in two steps. In the first step, the actual time spent in the different barn areas was calculated using location data. In the second step, accelerometer data were used to classify cow behaviour using the location information of step 1 (e.g., a cow located in the cubicles cannot be classified as feeding, or drinking). A total of 156 hours of video recordings were used for the validation. For each hour of data, the total time each cow spent in each area and performing which behaviours (feeding, drinking, ruminating, resting, and eating concentrates) were computed using the sensors and compared against annotated video recordings. Bland-Altman plots for the correlation and difference between the sensors and the video recording were then computed for the performance analysis. The overall performance of locating the animals into the correct functional areas was very high. The R was 0.99 (P < 0.001), and the root-mean-square error (RMSE) was 1.4 min (7.5% of the total time). The best performance was obtained for the feeding and lying areas (R = 0.99, P < 0.001). Performance was lower in the drinking area (R = 0.90, P < 0.01) and the concentrate feeder (R = 0.85, P < 0.05). For the combined location + accelerometer data, high overall performance (all behaviours) was obtained with an R of 0.99 (P < 0.001) and a RMSE of 1.6 min (12% of the total time). The combination of location and accelerometer data improved the RMSE of the feeding time and ruminating time compared to the accelerometer data alone (2.6-1.4 min). Moreover, the combination of location and accelerometer enabled accurate classification of additional behaviours that are difficult to detect using the accelerometer alone, such as eating concentrates and drinking (R = 0.85 and 0.90, respectively). This study demonstrates the potential of combining accelerometer and UWB location data for the design of a robust monitoring system for dairy cattle.