TY - JOUR

T1 - Time-series analysis for online recognition and localization of sick pig (sus scrofa) cough sounds

AU - Exadaktylos, Vasileios

AU - Silva, Mitchell

AU - Ferrari, Sara

AU - Guarino, Marcella

AU - Taylor, C. James

AU - Aerts, Jean-Marie

AU - Berckmans, Daniel

PY - 2008/12

Y1 - 2008/12

N2 - This paper considers the online localization of sick animals in pig houses. It presents an automated online recognition and localization procedure for sick pig cough sounds. The instantaneous energy of the signal is initially used to detect and extract individual sounds from a continuous recording and their duration is used as a pre–classifier. Auto–regression (AR) analysis is then employed to calculate an estimate of the sound signal and the parameters of the estimated signal are subsequently evaluated to identify the sick cough sounds. It is shown that the distribution of just 3 AR parameters provides an ade-quate classifier for sick pig coughs. A localization technique based on the time difference of arrival is evaluated on field data and is shown that it is of acceptable accuracy for this particular application. The algorithm is applied on continuous recordings from a pig house to evaluate its effectiveness. The correct identification ratio ranged from 73% (27% false positive identifications) to 93% (7% false positive identifications) depending on the position of the microphone that was used for the recording. Although the false negative identifications are about 50% it is shown that this accuracy can be enough for the purpose of this tool. Finally, it is suggested that the presented application can be used to online monitor the welfare in a pig house, and provide early diagnosis of a cough hazard and faster treatment of sick animals.

AB - This paper considers the online localization of sick animals in pig houses. It presents an automated online recognition and localization procedure for sick pig cough sounds. The instantaneous energy of the signal is initially used to detect and extract individual sounds from a continuous recording and their duration is used as a pre–classifier. Auto–regression (AR) analysis is then employed to calculate an estimate of the sound signal and the parameters of the estimated signal are subsequently evaluated to identify the sick cough sounds. It is shown that the distribution of just 3 AR parameters provides an ade-quate classifier for sick pig coughs. A localization technique based on the time difference of arrival is evaluated on field data and is shown that it is of acceptable accuracy for this particular application. The algorithm is applied on continuous recordings from a pig house to evaluate its effectiveness. The correct identification ratio ranged from 73% (27% false positive identifications) to 93% (7% false positive identifications) depending on the position of the microphone that was used for the recording. Although the false negative identifications are about 50% it is shown that this accuracy can be enough for the purpose of this tool. Finally, it is suggested that the presented application can be used to online monitor the welfare in a pig house, and provide early diagnosis of a cough hazard and faster treatment of sick animals.

KW - acoustic signal detection

KW - acoustic signal processing

KW - autoregressive processes

KW - bioacoustics

KW - diseases

KW - pattern classification

KW - time series

U2 - 10.1121/1.2998780

DO - 10.1121/1.2998780

M3 - Journal article

VL - 124

SP - 3803

EP - 3809

JO - Journal of the Acoustical Society of America

JF - Journal of the Acoustical Society of America

SN - 0001-4966

IS - 6

ER -