PSIII-6 Analysis of vocal sounds for detecting abnormal symptoms in sows

Abstract

Abstract Various technologies for animal health have been introduced and used in the livestock field as a part of an integrated processing methodology to construct a successful smart farm. This study aims to present a health assessment method applied to an individual pig using acoustic vibration. The experiment was based on the hypothesis that there is a strong relationship between acoustic phenotype and health condition. The information from a normal and abnormal sow was simultaneously and continuously recorded using a sound recorder for 24 hours. The abnormal sow was given an injection of 70% dextrose to the knee, which experienced necrosis due to a subsequent osmotic phenomenon. The experiment began at 9 am and continued until 8 am the next day and was repeated twice. During the experiment, the high-resolution recorder was located 50 cm from the top of the farrowing crate and directed at the sow’s head to reduce interferences from other sources of sound and noise. The first step of analysis was denoising the recorded acoustic information. Then, the Fourier fast transform was applied to the preprocessed data. Data were analyzed with PROC GLM (SAS 9.3), where a trial and treatment were included as fixed effects. The magnitude of frequency between normal and abnormal sows was significantly different (P < 0.05), in which the range of magnitude value was higher and lower than 0.015 for the normal and abnormal sow, respectively. The range 0.015 to 0.020 for the normal sow was clearly discriminated from the range 0.010 to 0.015 for the abnormal one. A more accurate interpretation of sows’ vocal data depends on the quality and quantity of data regarding their health condition. A promising algorithm of processing acoustic phenotype related to bio information could be useful in numerous complex health assessments.