Status Discrimination of Dairy Cows using Activity Meter and Machine Learning

摘要

It is important to detect the estrus of a cow in order to maintain lactation and increase farmer productivity. In the existing method, the observation interval is long, and it is possible to overlook the estrus. In this study, we propose a method of discriminating the condition of dairy cow using acceleration data. It is intended to detect changes of behaviors in dairy cows at shorter intervals and to enable precise estrus detection. We attached a small acceleration sensor to a cow and collected data. We extracted features from the collected data and applied machine learning to predict the cow's condition. In this study, we attached sensors to dairy cows at Arimura Farm and the Konsen Agricultural Experiment Station, and collected data. One dairy cow at each farm had a sensor attached. We extracted the feature quantity and discriminated the state and the 10- fold cross validation. It was possible to judge the state with an accuracy of 90% or more in collected data on both farms. In addition, when the collected data on both farms were combined and learned, it became possible to judge the state with an accuracy of 94.8%.