Research And Application On Multi-source Monitoring And Information Fusion Method For Porcine Abnormal Behaviors

As a vitally important part of agriculture,animal husbandry plays an unshakable role in the development of the national economy.Considering factors such as breeding environment,quality and efficiency,intensive pig breeding is of great significance.At present,modern intensive pig farms have gradually realized unattended.In order to further realize safe,high-quality and efficient breeding,it is imperative to intelligently monitor abnormal behaviors of pigs in unattended pig farms.Machine vision,voice recognition and ultrasonic technology are used comprehensively in this dissertation to monitor the abnormal behavior of pigs digitally from multiple angles.Finally,the method of multi-source information fusion is used to evaluate the abnormal behavior of pigs in multiple sources and time periods.The main research contents of this dissertation are as follows:(1)Monitoring of porcine abnormal behavior based on machine visionThe huddling behavior of pigs is closely related to disease prevention and comfort degree.The aggressive behavior can affect the health and welfare of pigs.In order to detect the huddling and aggressive behaviors of pigs in a complex environment,machine vision is introduced in this dissertation.In view of the limitations of traditional object detection methods in the detection of pigs in a complex breeding environment and the detection accuracy of closely contacted pigs,a pig detection method combined Signal Shot Multi Box Detector(SSD)and improved multi-view detection method is proposed.The dispersion of pigs is defined to represent the degree of pig aggregation.Through experimental comparison,the appropriate threshold of dispersion is set to detection the huddling behavior of pigs.In the process of detection of aggressive behavior,according to the characteristics of aggressive pigs,a moving pig detection method combined frame difference(FD)and SSD is proposed,which can effectively eliminate the interference of environmental factors and determine the position of pigs.Finally,the judgement method of aggressive behavior is designed according to the detected position information and the movement duration.It can effectively recognize the aggressive behaviors of pigs and provide judgment basis for breeders.(2)Recognition of porcine abnormal sounds based on improved MultiSVDDPigs will make different sounds when their physical condition changes or the external environment is stimulated.In this dissertation,the cough and scream are selected as porcine abnormal sounds.In order to detect and classify porcine cough and scream,the Multiple Support Vector Data Description(Multi-SVDD)is proposed.In view of the problem that,in the training of Multi-SVDD,the human errors on tagging training data can easily lead to underfitting,which affects the accuracy,Multi-SVDD is improved.According to the importance of each training sample,a corresponding weight is assigned to improve the error-tolerance of the model.The experimental results show that the improved Multi-SVDD can recognize porcine abnormal sounds effectively and it has strong applicability.(3)The judgment method for porcine abnormal diet based on improved PSOSVDDThe porcine diet can characterize the health of pigs to a certain extent.In this dissertation,the porcine diet data are collected by a pig diet data collection device based on ultrasonic designed by the research group.The SVDD is used to judge the porcine abnormal diet through the porcine daily diet times and diet time.Aiming at the problem that the penalty factor and kernel function parameters are difficult to determine in SVDD,the Particle Swarm Optimization(PSO)algorithm is used to optimize the SVDD parameters.Because of the fuzziness in the judgment of porcine abnormal diet,the accuracy of fuzzy decision by decision function is limited.Therefore,this dissertation constructs a fuzzy decision function for abnormal judgment.The improved PSO-SVDD is used to judge the porcine abnormal diet.The experimental results show that the method proposed in this dissertation can judge the dietary abnormality of pigs accurately.(4)The fusion evaluation of porcine Multi-source and Multi-period abnormal behaviors based on grey-evidential combination modelIn the process of monitoring abnormal behaviors of pigs,it is difficult for a single monitoring method to fully and accurately realize the automatic monitoring of abnormal behaviors.Therefore,it is necessary to make fusion judgments using different monitoring methods from multiple angles.At the same time,in order to avoid the contingency of abnormal judgment results in a signal period on the abnormal degree evaluation,the Grey-evidential combination model combined with grey clustering and D-S evidence theory is adopted in this dissertation.Aiming at the problem that D-S evidence theory cannot obtain reasonable results when fusing high conflict evidence,the confidence level of evidence is introduced to modify the combination rules.The method can effectively evaluate the porcine abnormal behaviors in multiple sources and periods.It can provide effective abnormal information of pigs for breeders and improve the monitoring efficiency.In this dissertation,an intelligent monitoring system for porcine abnormal behavior based on multi-source information fusion is designed for intensive pig farms.Machine vision,sound recognition and ultrasonic technology are used to detect multi-source abnormal information.And the multi-source abnormal information is evaluated comprehensively.It helps to improve the efficiency of intensive pig farms,and provides certain theoretical support and technical support for the upgrade and conversion of pig behavior monitoring from manual to digital and intelligent.