Study On Extraction And Recognition Method Of Characteristic Parameters Of Hogs’ Cough

In recent years,the economic development mode of the national pig breeding industry is developing towards intensive breeding.With the intensive development of the pig breeding industry,pig respiratory diseases have become one of the most common diseases in pig farms.Pig cough is an early symptom of respiratory disease,how to quickly determine whether the pig has respiratory diseases through the cough sound is one of the problems to be solved in the pig breeding industry.At present,the main method of observing pig cough sound is artificial observation,which not only consumes manpower and material resources,but also is not conducive to the growth of pigs,so the study of pig cough sound recognition detection system has great significance for the pig breeding industry.This paper introduces the technology and application of livestock and poultry voice recognition at home and abroad,and takes this as a reference to establish a cough sound recognition system for pigs from the construction of pig sound collection equipment,sample acquisition,pretreatment,characteristic parameter extraction and recognition algorithm,and takes the integration of feature parameters and the optimization of recognition algorithm as the research direction and breakthrough point to improve the cough sound recognition system of pig breeding,the main work includes:(1)Sound acquisition equipment construction and sample acquisition.After the field investigation of the pig farm environment,with the Raspberry Pi,7-inch HDMIips touch screen and four-microphone array as the core,the sound acquisition equipment was built to observe the sound collection in real time;under the guidance of the breeder,the Audacity audio processing software was used to intercept,manually classify and mark the collected sounds,and the pigs coughed,snor,snor,scream,and pig humming samples were obtained.(2)Sound pre-processing and feature parameter extraction.The resulting pig sound samples are pre-treated to ensure sample quality.The method of loading digital filter is used to preemphasize the pig sound to avoid the interference signal of the pig house and the pig sound signal required for the test to overlap;the FIR filter is used to denois the sound sample to avoid the impact of the environmental noise of the pig house on the test;the double limit gate method is used to detect the pig sound endpoint,locate the start and end point of the pig sound,and eliminate the sound of the silent segment.Analyzing the principle and advantages and disadvantages of shortterm energy,short-term average zero-crossing rate and Mel frequency reciprocal coefficient,and comparing the three sound characteristic parameters,the test shows that Mel frequency inversion coefficient is more able to combine the dynamic and static nature of sound,reflect the difference of pig cough sound,and improve the overall recognition rate.(3)Pig cough sound recognition model was established.In-depth study of the principle of artificial neural network,support vector machine and dynamic time regularity and related parameter selection,based on the Mel frequency inversion coefficient to construct artificial neural network,support vector machine and dynamic time regular pig cough sound recognition model.The performance evaluation of the three models was carried out by the method of 50% crossverification,and the average recognition rate was 74.33%,77.18%and 80.81%,respectively,compared with the other two models,the dynamic time regularization model had a better recognition effect.(4)Feature parameter fusion.The short-term energy and the Mel frequency reciprocal coefficient are fused into a new characteristic parameter STE-M by principal component analysis,and an artificial neural network,support vector machine and dynamic time-regular pig cough recognition model are constructed.The average recognition rates were 76.19%,80.20% and82.01%,respectively,and the results showed that STE-M could improve the overall performance of the pig cough recognition system.(5)Model series and parameter optimization.According to the first mock exam results of(3)and(4),the feasibility analysis of support vector machine and dynamic time warping algorithm is carried out in series.The output of time warping algorithm is used as input of support vector machine.STE-M is used to build DTW-SVM series model for sound characteristic parameters.The average recognition rate is 83.17%,higher than that of 1-3 percentage points of single model.The DTW-SVM series model is optimized,the particle swarm optimization algorithm is used to select the most appropriate penalty factor and kernel function width,and the PSODTW-SVM recognition model is established.The average recognition rate is 85.17%,which greatly improves the overall performance of the pig cough recognition system.