Research And Implementation Of Sheep Counting Algorithm Based On Computer Vision

Inner Mongolia sheep breeding industry occupies an important position in the grassland animal husbandry industry,sheep counting is an essential part of the livestock industry and are used by the government to monitor for overgrazing.It is particularly difficult to count the number of sheep due to the extensive grassland grazing areas,scene light changes,uneven distribution of sheep and serious occlusion among the sheep.So this paper proposes a dense counting method for grassland sheep based on computer vision technology to quickly and accurately count the number of sheep,which are important for guiding overgrazing monitoring and grassland ecology assessment.The research of this paper mainly includes the production of drone sheep dataset,the design of CNN-based density estimation counting model and the design of sheep dense counting application system in three parts.The first is the production of dataset,due to the lack of public sheep dataset,this paper makes a large-scale sheep dense counting dataset USC(UAV Sheep Counting)based on UAV images,containing 3510 images and 707257 sheep point annotations.Secondly,MCNN,CSRNet,SFANet,Bayesian Loss and other CNN-based density estimation crowd counting models were borrowed to conduct comparative experiments on the USC dataset and the experimental results were summarized by comparative analysis.Finally,based on the Bayesian Loss counting model,which has the best counting effect,a dense sheep counting application system is designed,which proceeds from two aspects of dense sheep counting algorithm and application system.A Web server is built on Ubuntu system,visualizing the data with Bootstrap and HTML5 on the frontend,and then the Flask framework is used to as the Web application development on the backend,while MySQL database technology is integrated to complete the design of the sheep counting application system,users can realize the counting of sheep through Web browser.In this paper,sheep counting method based on computer vision were proposed,which can accurately count sheep to provide decision support for the relevant departments in grassland management and grassland ecology protection.The made dataset USC provides a basic data support for the research and application of deep learning in this field;the designed sheep dense counting models give useful attempts in the field of sheep dense counting research and application,and provide a reference basis for grassland grazing management decisions;the designed sheep counting application system achieves the complete functionality of data transfer to the application,which can facilitate users to browse the counting system webpages on client terminal and execute counting.It is convenient,cost-saving and has a broader development prospects.In addition,it has great superiority and wide application value for counting the number of sheep in small and medium-sized pasture environments and is worth popularizing.