Calving and estrus detection in dairy cattle using a combination of indoor localization and accelerometer sensors

Accelerometers (neck- and leg-mounted) and ultra-wide band (UWB) indoor localization sensors were combined for the detection of calving and estrus in dairy cattle. In total, 13 pregnant cows and 12 cows with successful insemination were used in this study. Data were collected two weeks before and two weeks after delivery for calving. Similarly, data were collected two weeks before and two weeks after artificial insemination (AI) for estrus. Different cow variables were extracted from the raw data (e.g., lying time, number of steps, ruminating time, travelled distance) and used to build and test the detection models. Logistic regression models were developed for each individual sensor as well as for each combination of sensors (two or three) for both calving and estrus. Moreover, the detection performance within different time intervals (24 h, 12 h, 8 h, 4 h, and 2 h) before calving and AI was investigated. In general, for both calving and estrus, the performance of the detection within 2-4 h was lower than for 8 h24 h. However, the use of a combination of sensors increased the performance for all investigated detection time intervals. For calving, similar results were obtained for the detection within 24 h, 12 h, and 8 h. When one sensor was used for calving detection within 24-8 h, the localization sensor performed best (Precision (Pr) 73-77%, Sensitivity (Se) 57-58%, Area under curve (AUC) 90-91%), followed by the leg-mounted accelerometer (Pr 67-77%, Se 54-55%, AUC = 88-90%) and the neck-mounted accelerometer (Pr 50-53%, Se 47-48%, AUC = 86-88%). As for calving, the results of estrus were similar for the time intervals 24 h-8 h. In this case, similar results were obtained when using any of the three sensors separately as when combining a neck- and a leg-mounted accelerometers (Pr 86-89%, Se 73-77%). For both calving and estrus, the performance improved when localization was combined with either the neck- or leg-mounted accelerometer, especially for the sensitivity (73-91%). Finally, for the detection with one sensor within a time interval of 4 h or 2 h, the Pr and Se decreased to 55-65% and 42-62% for estrus and to 40-63% and 33-40% for calving. However, the combination of localization with either leg or neck-mounted accelerometer as well as the combination of the three sensors improved the Pr and Se compared to one sensor (Pr 72-87%, Se 63-85%). This study demonstrates the potential of combining different sensors in order to develop a multi-functional monitoring system for dairy cattle.