Recognition system for pig cough based on probabilistic neural networks

Until now the use of acoustic bio-responses in bio-environment control as i ndicators of animal-condition is limited to human perception. Coughing is a frequent symptom of many respiratory diseases affecting the airways and lu ngs of humans and animals. Registration of coughs from different pigs in a controlled test chamber was done in order to analyse the acoustical signal. A new approach is presente d to distinguish cough sounds from other sounds, such as grunts, metal clan ging and background noise, using neural networks as the classification meth od. Other signals (such as grunts, metal clanging, etc.) could also be dete cted. The best performance was obtained with a hybrid classifier that classifies coughs and metal clanging separately from the rest, giving better results c ompared to a probabilistic neural network (PNN) alone. The hybrid classifie r, which consists of a 2- and a 4-class PNN, gave high discrimination perfo rmance in the case of grunts, metal clanging and background noise (91.4, 63 .9 and 82.6%, respectively) and a performance of (91.9%) for correct classi fication in the case of coughs. (C) 2001 Silsoe Research Institute.