Design And Implementation Of Meishan Pig Continuous Cough Sound Monitoring System

With the scale and intensification of pig breeding,respiratory diseases have become one of the most common and serious porcine diseases in pig farms.Timely detection and early-warning of respiratory diseases can improve the prevention and treatment of porcine diseases,reduce the economic loss caused by respiratory diseases,and promote the healthy development of pig breeding.The conventional technique of porcine respiratory diseases early-warning is mainly based on human observation,which is time-consuming,laborious,subjective and less feasible.Meishan pig is a precious resource formed under the unique natural conditions in China.As one of the local breeding pigs,Meishan pig is famous for its high fertility.In recent years,many scholars have carried out effective research work in basic and applied research based on Meishan pigs.Acoustic analysis is an accurate and contactless signal processing method,which is widely used in the monitoring of animal physiological behavior.Combined with the current research status of porcine respiratory diseases,this research used acoustic analysis technology to develop a continuous cough sound monitoring system for Meishan pigs,and to provide an efficient information management platform for respiratory diseases early-warning in Meishan pigs.The main research contents of this project are as follows:(1)The research on sound signal pretreatment and feature extraction optimization of Meishan pigs.According to the analysis of Meishan pigs’ respiratory common clinical symptoms,experimental scheme was determined.The HD-B-1001 high-fidelity network pickup was used to collect Meishan pig sound signal,which was preprocessed by pre-weighting,windowing and framing,etc.According to the time domain characteristics such as amplitude and duration of the pre-processed sound signal,the spectrum subtraction was improved to increase the noise reduction performance.The starting and ending points of the non-silent segment from the sound signal were determined by double threshold endpoint detection method to obtain effective sound signal.The short time energy,the short time zero crossing rate,the resonance peak and MFCC cepstrum were extracted from the acquired non-silent sound signals.Then,MFCC features were optimized to improve their effectiveness.(2)Recognition method of continuous cough for Meishan pig based on machine learning.In order to solve the problem that traditional machine learning sample features require equal length input,DTW or interpolation method was used to make the features equal in length.Four classification algorithms were constructed as SVM,SRC,LDA and LS-SVM.The difference of the algorithm recognition performance between single features,dimensionally reduced features and combined features processed by DTW and interpolation method is comprehensively analyzed.The results show that for the single cough recognition of Meishan pigs,the combined features of MFCC and short-time energy E were optimized by LS-SVM classification algorithm when the interpolation isometrical length was processed to the three-fourths site,and the recognition accuracy is 86.77%.At the same time,a continuous cough recognition model based on LS-SVM was designed with the length distribution of the single cough interval in the continuous cough.(3)Study on the continuous cough recognition of Meishan pigs based on BLSTM-CTC.The continuous speech recognition technology was introduced into Meishan pig’s sound signal recognition to select the best combination of Meishan pig’s sound signal characteristics and construct the BLSTM-CTC acoustic model,which was used to identify Meishan pig’s continuous cough.The accuracy differences between LS-SVM model and BLSTM-CTC model under four evaluation indexes as model recognition accuracy,cough misrecognition rate,cough sound recall rate and continuous cough sound proportion were compared and analyzed.The results show that the BLSTM-CTC acoustic model was superior to the LS-SVM model on the whole,so the BLSTM-CTC acoustic model was selected to mining data for the Meishan pig continuous cough sound monitoring system.(4)Design and implementation of Meishan pig continuous cough sound monitoring software system.The middleware program to monitor,read and process Meishan pig’s sound signal is designed and implemented in Java.Meanwhile,the middleware can store the continuous cough detection result and the information of cough pigs in MySQL database.A monitoring system for continuous cough of Meishan pig(WEB side)was developed based on SpringBoot lightweight framework.The system presented the continuous cough detection results of Meishan pig,the administrator information of Meishan pig building and the information of Meishan pig,etc.With the stable operation of the system,the human-computer interaction interface was optimized to provide an efficient information system for monitoring and warning the respiratory diseases of Meishan pig.