Change in feeding behaviour is one of the indicators useful to help identifying when animals become ill. The need to analyse a large number of animals at a time due to the increase in the herd dimension in intensive farming has led to the use of automated systems. Among automated systems, inertial sensor-based systems have been utilised to distinguish behavioural patterns in livestock animals. In this study, a new approach based on statistical analyses of accelerometer data, which were collected from wearable sensors fixed at the cow's collar, was defined and developed in order to define thresholds suitable for real-time classification of cow feeding and standing behavioural activity. The obtained classifier could be implemented within a software tool of a movement sensor-based system composed of low-cost devices. Accuracy of the classification was assessed by computing specific indicators: Misclassification Rate, Sensitivity, Precision, Specificity, Quality Percentage, Branching Factor, and Miss Factor. The results showed that the classifier produced the following values of the indicators: 5.56%, 93.33%, 95.45%, 95.56%, 89.36%, 0.05, and 0.07, respectively. The proposed threshold-based classifier allows for monitoring individual cows automatically and continuously and it is suitable for Real Time Computing Applications, since it does not require high computational time and resources. © 2017 Elsevier B.V.
|Titolo:||Development of a threshold-based classifier for real-time recognition of cow feeding and standing behavioural activities from accelerometer data|
|Data di pubblicazione:||2017|
|Appare nelle tipologie:||1.1 Articolo in rivista|