| With the rapid development of China’s economy,people’s living standard has been greatly improved.The rapid growth in the demand for living nutrition such as meat,egg,milk,etc.has made the scale and intensification of livestock and poultry farming a mainstream and necessary farming model.In modern large-scale breeding,broilers have high incidence of respiratory diseases in all seasons,which is an important factor restricting the healthy development of broiler industry.In many kinds of broilers respiratory diseases,cough is a major clinical manifestation.At present,the monitoring of broiler respiratory diseases in broiler farming mainly depends on manual observation at night.By this way,the labor cost is high(time-consuming and laborious)and the real-time and effectiveness cannot be guaranteed.Therefore,accurate and real-time automatic detection of broiler coughs can curb the development of the epidemic as early as possible,reduce the stress caused by human observation,and improve broiler production efficiency.At the same time,it also avoids the hidden dangers of the outbreak of bird flu caused by human and poultry contact,which has certain practical application and reference value in the future with higher labor cost.In this paper,according to the needs of broiler farming and production,with white feather broilers as the research object,a respiratory disease monitoring system for white feather broilers with voice recognition technology as the key point was designed.It aims to monitor broiler cough efficiently and accurately which could realize intelligent and automatic early warning of broiler respiratory diseases.This research would improve the automation level of broiler breeding industry,and provide theoretical and technical support for its application.The main research contents are as follows:(1)The real-time perception system for broiler vocalization was designed.The HDB-1001 network pickup was used to build a real-time perception system for broiler vocalization based on network transmission.The system took the network pickup as the acquisition node,used the TCP/IP network transmission protocol to transmit the collected digital audio signal to the PC processing server,which could acquire the broiler sounds in a real-time,stably and automatically,thus provided effective and reliable data source for broiler respiratory disease monitoring system.It was the first key to realize automatic monitoring system.(2)The cough detection algorithm of broilers was researched.The vocalization data of white-feather broilers of different ages were obtained through several field broiler farm trials,and the vocalization database of white-feather broilers was established by manually intercepting the voice samples(cough,chirp,etc.)of white-feather broilers.After the noise measurement and analysis in the broiler house,the original audio preprocessing was carried out according to the frequency domain spectrum of different sound samples of broilers,and the sound enhancement(filtering and noise removal)method was mainly studied.Then,based on the characteristics of local time-frequency variation and non-stationary of broiler cough sound the Mel frequency cepstrum coefficient(WMFCC)feature of broiler sound samples based on wavelet transform was extracted.After designing and studying the two detection and classification models of GMM-HMM and keras multilayer perceptron,the extracted WMFCC feature parameters were fed into the model for training and experimental verification.The results showed that the detection efficiency of the GMM-HMM model was about 10 times faster than that of the broiler cough detection model based on multi-layer perceptron,and its detection accuracy rate was 95%on average after verification later,which meets the requirements of the detection system identification.(3)The application system for monitoring broiler respiratory disease was developed.The MySQL database,Java program and web page front-end visualization technology was use to build the application terminal system of broiler respiratory disease monitoring.Python was used to process the acquired sound data of broilers and feed it into the abnormal detection process.The detection results were saved in the database.The application-side system reads the detection results from the database in real time and displayed them on the web page.When the detection results exceeded the threshold set by the system,the application-side system would show an alarm to prompt the breeders to take corresponding measures.The management display of the application side followed the design principles of friendliness,effectiveness,and ease of use.So that breeders could be informed of abnormal health information during the growth of broilers in time,deal with emergencies effectively,and increase broiler farming productivity. |