The Hu-sheep Ruminant Behavior Recognition System Based On Acoustic Model And Machine Learning

At the help of Academia and industry attention to computer technology,artificial intelligence technology has been unprecedented progress,and made a great breakthrough.At present,artificial intelligence technology has been applied in various fields,speech recognition,natural language processing,image recognition have become the three main battlefields of artificial intelligence,the major internet companies and countless start-up companies are also actively preparing for artificial Smart related contest.Such as Baidu's in-depth study lab,ZTE's Artificial Intelligence Challenge.From a technical point of view,artificial intelligence technology has basically achieved practicality,but from the application of the actual situation it has not yet reached the real intelligence.Behind the rapid development of artificial intelligence is the machine learning algorithm innovation,and the machine learning algorithm will continue to apply to new areas to solve new problems.From the practical point of view,technology is powerful but it is only used to solve the needs of people and the society.Therefore,this study we are not on the artificial intelligence technology innovation,but concerned more about the machine learning base algorithms application.In the course of human history,animal husbandry has a history of thousands of years,in agriculture which has a pivotal position in recent years with the development of science and technology and large-scale breeding of the formation,making the animal husbandry more and more towards the direction of precision,While in the animal husbandry which ruminant livestock occupy half of the country,so the ruminant livestock research will promote the development of animal husbandry in China.Ruminant behavior as a unique feature of ruminant livestock can reflect the individual's physiological condition.At present,most of the ruminant behavior studies on ruminants rely on artificial observation.Artificial observation has its accuracy.However,the human and financial costs of direct observation are very high.Therefore,it is difficult to operate in large scale and accurate breeding.In view of the above problems,this paper studies,designs and realizes the Hu sheep rumination behavior recognition system based on acoustic model and machine learning.In this system,we propose a simple,practical and easy operation method to analyze and identify the ruminant of Hu sheep,in order to solve the ruminating behavior of Hu sheep and the classical classification algorithm of neural network and support vector machine in machine learning.First,we record the sound signals generated by the Hu sheep,to represent the ruminant event,and then through the preprocessing,candidate extraction and other steps to get the rumination of the candidate;after the feature extraction feature,followed by the neural network model,The support vector machine model is used to study and train the characteristics of these candidates,and then the new ruminating behavior candidate is identified as a ruminant event and a non-ruminant event through a trained model.In order to adequately test our proposed method,the test data used data containing less noise and more noise,respectively,since noise was more difficult to handle in all acoustic-based identification tasks,so we more concerned about the test data containing more noise,while the noise is more data closer to the actual situation.The experimental results show that the neural network model achieves an average accuracy of 86% in the statistics of Hu's ruminant count,while the support vector machine performs better on the small sample data and obtains an average accuracy of 90.4%.In addition,Rumination events and neural network models are lower,while the support vector machine model achieves an average matching rate of 90.1%and this effect is better than the foreign related research in ruminating behavior recognition.So,finally we decided to use SVM as the final classification algorithm in the system.At the same time,we designed a friendly interface for the system,making the operation very convenient,and the test run stable.