An online method for estimating grazing and rumination bouts using acoustic signals in grazing cattle

Abstract

The growth of the world population expected for the next decade will increase the demand for products derived from cattle (i.e., milk and meat). In this sense, precision livestock farming proposes to optimize livestock production using information and communication technologies for monitoring animals. Although there are several methodologies for monitoring foraging behavior, the acoustic method has shown to be successful in previous studies. However, there is no online acoustic method for the recognition of rumination and grazing bouts that can be implemented in a low-cost device. In this study, an online algorithm called bottom-up foraging activity recognizer (BUFAR) is proposed. The method is based on the recognition of jaw movements from sound, which are then analyzed by groups to recognize rumination and grazing bouts. Two variants of the activity recognizer were explored, which were based on a multilayer perceptron (BUFAR-MLP) and a decision tree (BUFAR-DT). These variants were evaluated and compared under the same conditions with a known method for offline analysis. Compared to the former method, the proposed method showed superior results in the estimation of grazing and rumination bouts. The MLP-variant showed the best results, reaching F1-scores higher than 0.75 for both activities. In addition, the MLP-variant outperformed a commercial rumination time estimation system. A great advantage of BUFAR is the low computational cost, which is about 50 times lower than that corresponding to the former method. The good performance and low computational cost makes BUFAR a highly feasible method for real-time execution in a low-cost embedded monitoring system. The advantages provided by this system will allow the development of a portable device for online monitoring of the foraging behavior of ruminants.