Among the automated systems used to monitor animal behaviour in real time, those based on wearable inertial sensors are widely used for dairy cows housed in free-stall barns. However, the technical specifications of these systems and the code of the implemented algorithms are seldom included in the literature. The overall aim of this study was to fill these gaps by proposing new open-source software tools, i.e. an algorithm and a classifier, to be adopted in a low-cost automated monitoring system based on accelerometers for discriminating dairy cow behavioural activities. Firstly, a novel algorithm characterised by a linear computational time was used for real-time monitoring and analysis of walking behaviour. An innovative classifier was then proposed to detect cow feeding and standing behavioural activities. Both these software tools were based on statistically defined thresholds computed from accelerometer data acquired during the animals’ daily routine by a new data acquisition system operating with a sampling frequency of 4 Hz. It required simple installation into the building and did not need any preliminary calibration. In this study, an overall algorithm for the recognition of dairy cows’ behavioural activities (i.e., lying, standing, walking, and feeding) was also proposed. With regard to the accuracy of the algorithm for walking, the total error was 9.5% and the relative measurement error ranged between 2.4% and 4.8%. The misclassification rate of the algorithm, which discriminates feeding from standing, was 5.56%. Testing of the whole algorithm for cow behaviour recognition showed relatively lower performance in discrimination of standing from walking. The performance of the data acquisition system was evaluated by the stored data index which achieved 83%. The application of the proposed tools allowed cow behaviour recognition with a high level of accuracy, using low-cost devices and open-source software.
Arcidiacono C., Porto S.M.C., Mancino M., Valenti F., Cascone G. (2017). New open-source software tools using accelerometer data for the discrimination of cow behavioural activities in free-stall barns. Nantes : European Conference on Precision Livestock Farming.
New open-source software tools using accelerometer data for the discrimination of cow behavioural activities in free-stall barns
Valenti F.;
2017
Abstract
Among the automated systems used to monitor animal behaviour in real time, those based on wearable inertial sensors are widely used for dairy cows housed in free-stall barns. However, the technical specifications of these systems and the code of the implemented algorithms are seldom included in the literature. The overall aim of this study was to fill these gaps by proposing new open-source software tools, i.e. an algorithm and a classifier, to be adopted in a low-cost automated monitoring system based on accelerometers for discriminating dairy cow behavioural activities. Firstly, a novel algorithm characterised by a linear computational time was used for real-time monitoring and analysis of walking behaviour. An innovative classifier was then proposed to detect cow feeding and standing behavioural activities. Both these software tools were based on statistically defined thresholds computed from accelerometer data acquired during the animals’ daily routine by a new data acquisition system operating with a sampling frequency of 4 Hz. It required simple installation into the building and did not need any preliminary calibration. In this study, an overall algorithm for the recognition of dairy cows’ behavioural activities (i.e., lying, standing, walking, and feeding) was also proposed. With regard to the accuracy of the algorithm for walking, the total error was 9.5% and the relative measurement error ranged between 2.4% and 4.8%. The misclassification rate of the algorithm, which discriminates feeding from standing, was 5.56%. Testing of the whole algorithm for cow behaviour recognition showed relatively lower performance in discrimination of standing from walking. The performance of the data acquisition system was evaluated by the stored data index which achieved 83%. The application of the proposed tools allowed cow behaviour recognition with a high level of accuracy, using low-cost devices and open-source software.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
