Acoustic Features for Pig Wasting Disease Detection

Failure to detect pig wasting disease in a timely and accurate manner can become a limiting factor in achieving efficient reproductive performance. In this paper, we discuss the methodology and results of a study we conducted to analyze for acoustic features for pig wasting disease and to use pattern recognition to examine the differences of the sounds pigs make when affected with pig wasting disease compared to normal sounds. The proposed acoustic feature subset selection algorithm indicated that {F7, RMS, Max Pitch, PSD1, Peak frequency} is the optimal feature subset that can be used to detect pig wasting disease. Finally, the results of the performance evaluation conducted using real sound data from an audio surveillance system showed that the average detection accuracy approached 98.4%, with FPR and FNR reaching on average 0.2% and 1.6%, respectively, when a support vector machine algorithm was used for detection. Moreover, we empirically confirmed that the performance of the optimal acoustic feature subset does not depend on the specific detector used.