Fuzzy-neural approach for improved classification of pig-cough frequency features

A. van Hirtum, A. Chedad and D. Berckmans

Laboratory for Agricultural Buildings Research, University of Leuven, Heverlee, Belgium

The ability to adapt information of acoustic bio-responses from pigs in animal houses will mainly depend on an accurate automatic recognition of the sound of interest. A natural acoustic indicator of animal welfare is the appearance (or absence) of coughing in the animal habitat. In previous research a simple on-line recognition technique for pig coughing has been presented. This resulted in positive cough recognition of 92% of the sound database consisting of 5319 individual sounds among which 2034 coughing. Sounds were collected on 6 healthy animals and contained both animal vocalisations and background noises. Each of the test animals was repeatedly placed in a laboratory installation in which coughing was induced by evaporation of citric acid. A 2-class classification into 'cough' or 'other' was done by application of a distance function to a spectral sound analysis. However, for the whole sound database there was a misclassification of 21%. As spectral information up to 10000 Hz is available in this paper an improved overall classification on the same database is presented by applying the distance function to several frequency ranges in combination with a neural network approach.


Poster presented at Measuring Behavior 2000, 3rd International Conference on Methods and Techniques in Behavioral Research, 15-18 August 2000, Nijmegen, The Netherlands

© 2000 Noldus Information Technology b.v.