Identification And Tracking Of Livestock Based On Deep Learning In Intelligent Farming

Recently,people’s demand for meat food is getting bigger and bigger,but the relatively backward breeding methods of farms restrict the production capacity of meat food,and the problems of labor shortage and low breeding efficiency are prominent.Promoting the informatization and intelligence of farms can solve these problems and increase the output of farms.Due to the reducing of the cost of monitoring equipment,farms can install cameras on a large scale for monitoring,which can also generate a large amount of video data while ensuring safety.The combination of big data and computer vision technology can help researchers design intelligent algorithms to help staff manage farms efficiently.Thus,based on the surveillance images of livestock farms,the artificial intelligence technology will develop efficient intelligent algorithms to promote the development of intelligent breeding.Based on surveillance images,three problems are studied mainly in the paper:intelligent identification algorithm,intelligent tracking algorithm and intelligent behavior recognition algorithm for livestock.Intelligent recognition,tracking and behavior recognition algorithms for livestock,involve target detection technology,multi-target tracking technology and behavior recognition technology in the field of computer vision respectively.The monitoring image data of the farm has problems such as sample imbalance,difficulty in identifying small targets,unbalanced lighting,serious occlusion,and low sample differentiation,which have a great impact on the performance of the algorithm.In this paper,we design and improve the algorithm for the difficulties of the data set.For the intelligent recognition algorithm of livestock,high-resolution network and other technical means are proposed to improve the information extraction and expression ability of the network;For the intelligent tracking algorithm of livestock,a network based on graph theory is proposed,which transforms the tracking problem into a graph update and node matching problem,and makes full use of the spatial information and texture information of the image;For the intelligent behavior recognition algorithm of livestock,an R-C3D algorithm combined with The Antenna network is proposed.It combines the the intelligent identification algorithm and the intelligent tracking algorithm of livestock to complete the behavior recognition task of dense targets.The algorithms proposed can achieve good results on the livestock dataset,and the ablation experiments designed have proved that the proposed improvements are effective.The research of this paper has achieved good results in livestock surveillance images.The livestock intelligent identification algorithm reached 58.38%of the mmAP value,the livestock intelligent tracking algorithm reached 78.4%of the MOTA value,and the livestock intelligent behavior recognition algorithm reached 74.3%of the mAP value.This shows that the research of this paper has certain research significance and practical application value in the field of intelligent identification,tracking and behavior recognition of livestock based on surveillance images。...