Recognition Method Of Pig Cough Sounds Based On Voice Recognition Technology

Pig breeding is the pillar industry of animal husbandry in China.The continuous development of science and technology promotes the transformation of pig breeding from traditional extensive type,family type to scale,automation and intelligent culture.The problems of small raising space,large density and poor ventilation have aggravated the risk of respiratory diseases in pigs,and how to quickly monitor and warn the respiratory diseases of pigs is an urgent problem to be solved to expand the scale of the industry.At present,artificial observation of the cough situation of pigs is the main method to evaluate the infection of respiratory diseases.However,this method is time-consuming and hard work and subjective,and human activity is easy to cause stress response of pigs and is not conducive to the growth of pigs.Therefore,developing automatic identification and monitoring system for pig cough has great research value and broad market prospects.Voice recognition technology has the characteristics of fast response,low cost,objectivity,accuracy,non-contact and non-invasive.This paper collects the sound and noise of pig's cough in the piggery environment,obtains experimental samples by artificial markers,preprocessing using pre accentuation and spectral subtraction de-noising,and double threshold end point detection to extract short time energy(EN),short-time zero crossing rate(ZCR)and improved Mel cepstrum coefficient(MFCC).Finally,we construct vector quantization(VQ),support vector machine(SVM)and Hidden Markov Model(HMM),and optimize the pig cough sound recognition system,and explore a new way to realize automatic recognition of pig cough by sound recognition technology.The contents and results of this paper are as follows:1)The pig sound signals under the actual piggery were collected,and the samples were artificially marked and classified by software.Pig's cough and Five kinds of noise,such as eating sound and ear throwing,humming,screaming and sneezing,were obtained.and the voice was pre accentated,LMS algorithm and spectral subtraction to reduce noise.The quality of the test samples was guaranteed.2)The principle and advantages and disadvantages of short time energy,short-time zero crossing rate,Mel cepstrum coefficient and improved Mel cepstrum coefficient are analyzed,and endpoint detection is carried out with short time zero crossing rate and short time energy setting double threshold,which accurately locates the starting point of pig sound sample.Comparing the standard MFCC with the improved MFCC characteristic parameters,the experiment shows that the improved MFCC combines the static and dynamic characteristics,which can better reflect the specificity of the cough and improve the overall recognition rate.3)The principle of VQ,SVM and HMM and the selection of related parameters are studied.Based on the characteristic parameters of improved MFCC and short-time energy and improved MFCC fusion,the model of VQ,SVM and HMM pig cough recognition is constructed.The performance of the half off cross validation model was used to evaluate the recognition performance of the three models through the rate of cough recognition,non cough recognition rate,comprehensive recognition rate,and average recognition rate of the three models.The experimental results show that the average recognition rate of VQ,SVM and HMM based on the fusion feature parameters is increased from 75.18%,85.11%,and 89.38% to 81.64%,89.04% and 90.17%,compared with the three recognition models constructed with a separate improved MFCC.It shows that proper feature parameter fusion can improve the sound of pig's cough.The recognition performance of the system.4)The possibility and advantage of the fusion of HMM and SVM algorithm are analyzed.Through the series fusion,the output of HMM is used as the input of SVM,and the HMM-SVM fusion model is established.The average recognition rate is 91.59%.It shows that the series fusion of HMM and SVM is helpful to improve the performance of the recognition system.?5)Particle swarm optimization(PSO)is applied to optimize the HMM-SVM fusion model,and the recognition rate is increased to 92.79%.Four recognition models of VQ,SVM,HMM and HMM-SVM-PSO based on fusion characteristic parameters are used as base classifiers,and four base classifiers are fused by weighted voting algorithm.The performance of the pig cough recognition system is greatly improved,and the average recognition rate is 94.21%.