A pattern recognition approach for detecting and classifying jaw movements in grazing cattle

Abstract

Precision livestock farming is a multidisciplinary science that aims to manage individual animals by continuous real-time monitoring their health and welfare. Estimation of forage intake and monitoring the feeding behavior are key activities to evaluate the health and welfare state of animals. Acoustic monitoring is a practical way of performing these tasks, however it is a difficult task because masticatory events (bite, chew and chew-bite) must be detected and classified in real-time from signals acquired in noisy environments. Acoustic-based algorithms have shown promising results, however they were limited by the effects of noises, the simplicity of classification rules, or the computational cost. In this work, a new algorithm called Chew-Bite Intelligent Algorithm (CBIA) is proposed using concepts and tools derived from pattern recognition and machine learning areas. It includes (i) a signal conditioning stage to attenuate the effects of noises and trends, (ii) a pre-processing stage to reduce the overall computational cost, (iii) an improved set of features to characterize jaw-movements, and (iv) a machine learning model to improve the discrimination capabilities of the algorithm. Three signal conditioning techniques and six machine learning models are evaluated. The overall performance is assessed on two independent data sets, using metrics like recognition rate, recall, precision and computational cost. The results demonstrate that CBIA achieves a 90% recognition rate with a marginal increment of computational cost. Compared with state-of-the-art algorithms, CBIA improves the recognition rate by 10%, even in difficult scenarios.