Automatic recognition of ingestive sounds of cattle based on hidden Markov models

MILONE, D. H.; PADRÓN, M. S.; GALLI, J. R.; CANGIANO, C. A.; RUFINER, H. L.

Santa Fe, Argentina

Conferencia; XXXIV Conferencia Latinoamericana de Informática; 2008

An automatic tool capable of informing when bites and chews occurred, using only the acoustic signal, is presented. This information is useful for the estimation of intake and the monitoring of grazing behaviour. Additionally, the sound recording is easy to implement through the placement of a wireless microphone in the forehead of the animal, without altering its normal behaviour. Hidden Markov models have been used to segment and classify acoustic signals with good results. In this work we propose the use of hidden Markov models to recognize masticatory sounds of cattle. Three types of masticatory events from cows grazing pastures of two plant species (alfalfa and fescue) and two heights (tall and short) were successfully recognized. Recognition rates were 83.95% for tall alfalfa, 65.33% for short alfalfa, 84.68% for tall fescue and 83.68% for short fescue.