Pig Anomaly Detection Based On Audio Analysis Rechnology

In the aquaculture industry,the health status,welfare level and reproductive efficiency of livestock are the important indexes to judge the breeding technology and the main factors that determine the income of farming.The most important is the health of livestock,once the epidemic disease in the farm breeding animals broke out,not only will affect the level of the farm harvest,but also the entire aquaculture industry may also brought huge economic losses,and the community caused by bad influences.At present,the judgement of the health and welfare level of the livestock is mainly stay in the breeder observation.Keeping the staff with less and different levels of experience,resulting in uneven levels of judgment,and inevitably missed the phenomenon.So it is difficult to meet the needs of both the scale and the safety assurance.Pig industry is an important industry in China's agriculture,to ensure the safety of meat supply has an crucial role.With the continuous development of digital monitoring technology,the use of remote monitoring technology instead of artificial detection of pig health has become the focus of the current research.Based on the analysis of the monitoring system and the development of speech recognition home and abroad,this paper puts forward “Pig Anomaly Detection Based on Audio Analysis Rechnology”,which can improve the management efficiency of the farm and improve the health and welfare of the pig.This paper first introduced the existing digital analysis technology of livestock home and abroad,and selected the audio analysis technology as the research direction and the breakthrough point among the audio analysis technology,the machine vision technology,the wireless sensor technology,the radio frequency identification technology and the ultrasonic technology.It recongnized pig cough sound,pig fighting sound,pig hungry sound and pig convulsion sound by the voice recognition technology.The research work of this paper mainly includes:The application of integrated monitoring system based on streaming media technology in farms.After studying the overall structure of the integrated monitoring system,this paper analyzes the hardware(pickup,camera,etc.)and software(iVMS-4200)needed by the monitoring system;Then it analyzed the coding technology of FFmpeg to audio,and use FFepeg to re-encode the date collected by the integrated monitoring system Data from the "composite stream" to audio data;Moreover,it studyed the Spectral subtraction noise reduction,the Wavelet threshold noise reduction,and the Adaptive filtering cancellation.Finally,it analyzed the noise reduction principle of three methods and compared with the noise reduction effect of three kinds,as a basis for the selection of a noise which reduction effect is the best and provide higher Signal to Noise Ratio signal for the late voice recognition.The study on characteristic parameters extraction and recognition of pigcough sound.The acoustic characteristics of pig cough sound were analyzed and the characteristic parameters of it were extracted.Moreover,the recognition of pig cough sound was realized by using Hidden Markov Model.Firstly,pre-emphasis,sub-frame windowing and endpoint detection are used to preprocess the cough.Then,the characteristic parameters are extracted for each frame's speech signal,and the hidden parameter is input into the hidden Markov model with the characteristic parameter sequence.Finally,through the Baum-Welch algorithm to train the model,using the trained model to identify the pig cough sound by Vierbi algorithm.In addition,the characteristic parameters include Linear Prediction Cepstral Coefficient,the Mel-Frequency Cepstral Coefficients and their first order differences.The experimental results show that the hidden Markov model can complete the recognition of pig cough,and the recognition rate is the highest when the Mel-Frequency Cepstral Coefficients and its first order difference are the characteristic parameters.Four kinds of abnormal sounds of pig are analyzed,and the four kinds of abnormal sounds are identified by Vector Quantization,Dynamic Time Warping algorithm and Hidden Markov Model respectively.On the basis of these three kinds of recognition methods are optimized by some kind of algorithm.In this paper,the characteristics and differences of four abnormal pig sounds,such as pig fighting sound,hungry sound,cough sound and convulsion sound,are analyzed from the time domain,frequency domain and transform domain respectively.Moreover,two other methods of speech recognition areintroduced:Vector Quantization and Dynamic Time Warping algorithm.The characteristics of the three recognition methods and the recognition results of the abnormal sound of the pig are analyzed.Apart from this,the "Fusion classification algorithm" is proposed to optimize the three speech recognition methods.The experimental results show that the algorithm can be applied to the optimization of three models,and the recognition rate of pig abnormalities is improved.