Classification and correlation of coughing sounds and disease status in fattening pigs

There have been attempts to use technology to distinguish pig coughs from other sounds on farms. Machine learning is being used to classify pig coughs via python. Sound files have been converted to images that are composed of wave plots,spectrograms and log power spectrograms for identification of those sounds. A recorder was used with a total of 45 healthy three-bred weaned piglets, wherein three replications of each were used with 15 weaning pigs per pen during different months. This process was set up within the housing unit at a ratio as 1:1 (recorder per pen). Sounds, blood samples and tonsil swabs were collected every month. Pig cough sounds were then classified from other sounds and a coughing index (CI) was established. Blood samples and tonsil swabs were utilized to determine respiratory diseases via laboratory tests that included ELISA, PCR and bacterial cultures. According to our results, pig coughs sound distinctly different from other sounds as had been classified by python. Moreover, the laboratory results of the seroprofile of Mycoplasma hyopneumoniae (M.hyo), Porcine Reproductive and Respiratory Syndrome virus (PRRSv), and Porcine Circovirus type 2 (PCV2), as was established by ELISA test, were employed in disease detection during the fattening period. Spearman rank correlations and Kappa analysis were used to establish correlation values between coughing and the results of laboratory tests. CI revealed a high correlation coefficient and agreement with the ELISA results of M.hyo, as well as the PCR results of PRRSv and PCV2 (p<0.05), while CI also revealed low correlation coefficient and agreement with the results of the Streptococcus spp. and Pasteurella spp. cultures (p>0.05). Therefore, the monitoring of coughing can be suited to detect respiratory problems and any potential relationships with M.hyo, PRRSv and PCV2 infections