Subsequence clustering algorithm based on structural similarity and its application in cow estrus detection

The efficient and accurate estrus detection technology can enhance conception rates and shorten birth spacing of the breeding dairy cow, which is an important means for improving economic benefit. For the large-scale, intensive farming environment, the cow's behavior and activity is an important indicator to determine whether estrus which is confirmed by numerous academic and scientific research. Usually the cow's behavior decision-making methods are using the single point of data to classify behavior. However the cow's locomotory sensor data were multivariate time series data which were collected by time sequence. This paper presented a subsequence fast clustering algorithm based on structural similarity (SC-SS). The algorithm first partitioned the sensing time series data into several subsequence segment according to the first-order differential value of acceleration; then calculating the structure similarity of each subsequence segment by comparing their features of acceleration, energy, standard deviation; Finally the subsequence were grouped into three clusters. Experiment results on real data set demonstrate that the SC-SS is more efficient than K-means, has and more effective for classification of cow's behavior, which can increase the accuracy of cow's estrus detection.