Research On Horse Image Segmentation System Based On Computer Vision

In recent years,deep learning technology has been widely used in computer vision.More and more application scenarios in real life need relevant knowledge or semantics,which is inferred from images for further analysis and understanding.Semantic segmentation is the basis of scene understanding.Image analysis technology based on such technology has gradually penetrated into various fields of livestock management.How to obtain horse information through images has become an important technical node for processing image information by computer technology.The accuracy of target horse segmentation really affects further researches,such as horse behavior recognition,horse height measurement,horse weight measurement,and body health analysis.The result of image segmentation determines the reliability of subsequent work.This paper focus on the key issues encountered in horse image segmentation,such as unclear segmentation edges and poor segmentation results,and proposed a horse image segmentation algorithm based on fully convolutional neural network and fully connected conditional random field.The main research tasks as follows:(1)Build horse image segmentation dataset.The images are collected at Zhaosu Horsefarm in Yili.The shooting time covers morning,noon and evening,and original data contains strong light and shadow noise.The transfer learning is performed on the existed deep convolutional neural network(DCNN),and the fully connected conditional random field(CRF)is added to improve segmentation edge.Compared with the original model,the results show that Intersection over Union(IOU)has increased 3.8%in this method.(2)Compared the current popular image segmentation algorithms DCNN with MASKRCNN and FCN,analyze the advantages and disadvantages of each algorithm in terms of segmentation accuracy,intersection over union,and model consumption time.Transfer learning based on the original model.Analyze the segmentation accuracy.(3)Based on the work above,the horse image segmentation algorithm is applied to the practical application-horse image segmentation system which uses the open source Django as the basic framework,the front-end development uses HTML5,CSS,JavaScript technology,the database uses SQLite,the basic programming language is python3.The service implementation contains horse image segmentation on the web side.The system lays the foundation for horse analysis and scene understanding.As a sub-function of the XinJiang horse industry horse production process management platform,this system implements the segmentation function of horse registration pictures and the horse electronic medical record pictures,and provides a research basis for body measurement prediction and disease analysis.