3d acceleration for heat detection in dairy cows

Abstract

Accurate and reliable detection of heat in dairy cows is essential for a controlled reproduction and therefore, for maintaining milk production. Classical approaches like visual identification are no longer viable on large dairy herds. Several automated techniques of detection have been proposed, but expected results are only achieved by expensive or invasive methods, because practical methods are not reliable. We present a method that aims to be both practical and accurate. It is based on simple attributes extracted from 3D acceleration data and well known classifiers: multilayer perceptrons, support vector machines and decision trees. Results show promising detection ratios, above 90% in several configurations of the detection system. Best results are achieved with multilayer perceptrons. This information could be readily incorporated to the automated system in a dairy farm and help to improve its efficiency.