Evaluation of an Algorithm for Cough Detection in Pig Houses

Coughing is one of the most frequent presenting symptoms of many diseases affecting the airways and the lungs of both humans and animals. In piggeries, the continuous on-line monitoring of cough sound can be used to build an intelligent alarm system for the early detection of diseases. In this study, a previously developed algorithm was tested in field conditions. Pig cough sounds were registered on 44pigs by an operator holding a microphone at about 20 to 50 cm from the pigs head. From these sound files, feature vectors were extracted, containing information on the sound energy, spectral properties and time derivates. These feature vectors were compared to a reference set by means of a dynamic time warping algorithm. This leads to a two class classification: ill, no ill. The classification was checked by a veterinarian and found to be correct in 86 % of the cases.