ZigBee-based wireless sensor networks for classifying the behaviour of a herd of animals using classification trees

An in-depth study of wireless sensor networks applied to the monitoring of animal behaviour in the field is described. Herd motion data, such as the pitch angle of the neck and movement velocity, were monitored by an MTS310 sensor board equipped with a 2-axis accelerometer and received signal strength indicator functionality in a single-hop wireless sensor network. Pitch angle measurements and velocity estimates were transmitted through a wireless sensor network based on the ZigBee communication protocol. After data filtering, the pitch angle measurements together with velocity estimates were used to classify the animal behaviour into two classes; as activity and inactivity. Considering all the advantages and drawbacks of classification trees compared to neural network and fuzzy logic classifiers a general classification tree was preferred. The classification tree was constructed based on the measurements of the pitch angle of the neck and movement velocity of some animals in the herd and was used to predict the behaviour of other animals in the herd. The results showed that there was a large improvement in the classification accuracy if both the pitch angle of the neck and the velocity were employed as predictors when compared to just pitch angle or just velocity employed as a single predictor. The classification results showed the possibility of determining a general decision rule which can classify the behaviour of each individual in a herd of animals. The results were confirmed by manual registration and by GPS measurements.