Research On Image Segmentation And Behavior Recognition Of Pigs In Large-scale Farmin

In recent years,the number of pigs has continued to grow as a whole.Even under the influence of the ’COVID-19’,the global production of pigs has reached several billions.However,the traditional method of relying on manual pig monitoring has some problems,such as time and labor consuming,unstable subjective factors and low timeliness.Therefore,this paper combines pig breeding with modern monitoring methods,and conducts research on individual segmentation and behavior recognition of large-scale pig breeding.It aims to reduce the burden of farmers,improve animal welfare and provide solutions for the continuous automatic monitoring of individual animals.Aiming at the research on the image segmentation algorithm of pigs,in view of the fact that the traditional image segmentation algorithms are easy to be affected by light,noise,environment,etc.At the same time,the current pig segmentation algorithm has certain limitations in the number of pigs,pig size,pig breeding background and light environment.Therefore,this paper uses the Faster R-CNN model to realize image segmentation to obtain pig targets;In order to further obtain the position and area information of the pigs,and realize the instance segmentation of the pig images,the improved Mask R-CNN algorithm is adopted.The experimental results show that the two segmentation algorithms can realize the image segmentation of pigs and the acquisition of individual pig.The model has strong generalization ability and can be suitable for the actual pig breeding environment,which shows that the effectiveness of the pig segmentation algorithm in this paper.In particular,the improved Mask R-CNN network model can realize the instance segmentation of pigs,and can describe the edge of pigs very well.Aiming at the research on the behavior recognition algorithm of pigs,in order to balance the needs of the behavior detection speed and detection accuracy,the latest version of the YOLO detection algorithm(YOLOv5)is used to recognize the behavior of the pigs,and finally realize the three daily behaviors of pigs(standing,lie＿down and lie＿side).In order to solve the problem of training and learning the redundant information of pigs by the horizontal bounding box in the YOLOv5 and the overlapping of the borders when pigs gather,this paper proposes the YOLOv5＿rotate algorithm,which uses the rotating bounding box to extract the effective pig features and avoid the probability of missing detectione.Finally,using Precision,Recall,m AP@0.5,the P-R curve to analyze and compare the two models,which proves that the algorithm can frame pig targets more accurately on the basis of behavior recognition.In order to facilitate subsequent applications,enable users to intuitively observe the results of pigs segmentation and behavior recognition,and promote the intelligent development of pig breeding,this paper builds a pig analysis system based on the Web.The system uses the SpringBoot+Vue framework to realize the functions of pig analysis,historical data query and so on.It completes the visual monitoring and analysis of pigs,which can be applied to the intelligent scenes of large-scale breeding in the future.