Research On Cattle Muzzle Recognition Method Based On Machine Vision

The development of animal husbandry pays more and more attention to quality,traceability,and batch management.To improve the problems of traditional individual cattle recognition,a method of cattle identification based on machine vision is proposed.This study uses the deep learning method to extract the cattle muzzle image.To use the feature extraction algorithm recognition of the cattle muzzle image,which has a good recognition effect and has guiding significance for non-contact cattle identity recognition.This study consists of three parts: dataset creation of cattle muzzle images,target detection in cattle muzzle region and design of cattle muzzle image recognition model.Firstly,use a camera to take the cattle muzzle image and make the cattle muzzle image data set(CMID)at the cattle farms and milk stations in Inner Mongolia.The data set contains184 tested cattle and 11509 pictures.Then this study uses the Aerial Detection target detection model to detect the target of cattle muzzle images and obtain the image containing only cattle muzzle images.Using the uniform resolution and contrast limited adaptive histogram equalization(CLAHE)preprocessed the cattle muzzle image to enhance the recognizability.This study use experiments to compare and analyze the recognition effects of two feature extraction algorithms based on local invariant features and deep learning.Scale-invariant feature transform(SIFT),speeded-up robust features(SURF),and oriented fast and rotated BRIEF(ORB)based on local invariant features are used to extract the feature descriptors of cattle muzzle image.The K-Nearest Neighbor(KNN)classification algorithm matches the feature descriptors of the cattle muzzle image.This study uses the convolution neural network based on deep learning to extract the cattle muzzle image features and compare the effects of different image processing algorithms on the recognition effect.The experimental results show that CLAHE image preprocessing and SIFT best affect cattle muzzle image recognition in the one-to-many cattle identity retrieval experiment.The convolution neural network recognition effect is good in the one-to-one cattle identity matching experiment.In this paper,adopting the machine vision method,which does not influence individual cattle,achieves a good recognition effect for the muzzle image of livestock.It makes up for the shortcomings of traditional individual cattle identification and impacts the non-contact recognition of individual cattle.