Porcine Abnormal Sound Recognition And Monitoring System Based On Support Vector Machine

China is a large country in the world of pork production,and also a large country in the world of pork consumption,the breeding of porcine in China plays a very important role in animal husbandry.With the development of science and technology,more and more new technologies are applied to the porcine breeding process,which promotes the transformation of porcine breeding methods from extensive and individual free-range to intensive,large-scale and automated farming.Although intensive farming methods outperform traditional farming methods in terms of breeding efficiency and resource utilization,intensive farming tends to cause crowded living space,poor ventilation,and rapid spread of infectious diseases,which would lead to an increased probability of abnormal pigs.Therefore,how to timely warn the abnormal conditions of pigs is a major problem that needs to be solved in intensive porcine breeding.At present,the domestic porcine farming is mainly based on artificial monitoring,this method is not only time-consuming and laborious,but also vulnerable to subjective factors,at the same time,human activities would cause stress response of pigs,which is not conducive to the healthy growth of pigs.Sound recognition technology is widely used in security system,with highrecognition accuracy,no contact,strong objectivity and other characteristics.In this paper,a porcine abnormal sound monitoring system based on sound recognition technology is designed to monitor the frightened and cough of pigs on-line.The main contents and results of this paper are as follows:(1)When the sound signal in piggery is collected online,there are a large number of invalid sound signals with low energy amplitude in the collected sound signal.In this paper,energy acquisition threshold is set to reduce the processing of invalid sound signals and improve the speed of sound processing in piggery by monitoring system.(2)The sound signals of pigs collected in piggery are mixed with a large amount of environmental noise,mainly the noise of ventilators in piggery.In this paper,digital filter,wavelet threshold method and improved spectral subtraction method are used to denoise the collected porcine sound signals.The experimental results show that when the porcine sound signals are disturbed by fan noise,the improved spectral subtraction method can achieve better denoising effect,and can also meet the requirements of online processing.(3)Extracting four kinds of characteristic parameters for the porcine sound signal,including time domain characteristic parameters,LPCC characteristic parameters,MFCC characteristic parameters and feature parameters based on the elimination of slow changing trend items.These four characteristic parameters are combined with each other and the experiment is carried out.The results show that when the MFCC and its first-order differential coefficient areused as characteristic parameters,the abnormal sound recognition effect of porcine is the best.(4)Using the single-class detection capability of the SVDD model,and constructing a suitable multi-classifier for the SVM model,the detection and recognition models of SVDD and SVM are obtained.The model is used to detect and identify the abnormal sound signals of the porcine in the piggery.And the health status of the porcine is judged based on the final recognition results over a period of time.(5)The upper computer interface of the porcine abnormal sound monitoring system is designed by using Siemens configuration software WinCC7.0.It can display,record and store the changing trend of abnormal sound data in piggery,the processing result and alarm information in real time,and provide help for the feeders to know the situation in piggery in time.The abnormal porcine sound monitoring system designed in this paper can monitor the health status of pigs in piggery in real time,and provide a basis for realizing the abnormal monitoring of pigs from manual mode to automatic and digital mode.