Improving Oestrus Detection in Dairy Cows by Combining Statistical Detection with Fuzzy Logic Classification

Ecient automated oestrus detection in cows and heifers deeply influences reproductive performance of the animals, and the livestock farmers' protability. The main problem for practical application of automated detection is the high number generation of false-positive alerts. False alerts could be triggered by changes in feeding or heard behaviour. The detection to false alarm ratio need be very high to get farmers' condence in an oestrus detection system. Therefore, a method to enhance detection and reduce false alarm probabilities is necessary. Earlier research investigated statistical change detection and hypothesis testing applied on activity sensor data. This paper enhances earlier method by employing fuzzy logic technique to classify oestrus alerts from a model-based detection method utilising the cyclic nature of oestrus. Based on the distribution of the trait period since last detected oestrus, a set of membership functions is introduced with the objective of decreasing the number of false positive alerts as well as improve missed detection rate. The approach was tested on data from twelve diary cows collected over six months. The results show that the number of true detected cases decreased slightly after classication but false positive alerts were almost eliminated.