Research On Cattle Individual Identification Based On Cattle Face Image

In recent years,with the combination of science and technology and Internet technology,the animal husbandry industry has been promoted to undergo great changes,and new farming models such as smart agriculture and smart ranch have emerged,and the identification of individual cattle as the core content of the new farming methods has received a lot of attention.In this paper,we propose a cattle identification technology based on cattle face images using a non-contact identification method,which shows a better performance for the scenario of cattle insurance industry and is worthy of application and promotion.The research of this paper mainly includes four parts:acquisition and production of dataset,target detection of cattle face based on YOLOv5,cattle face feature extraction based on Inception structure and design and implementation of cattle face recognition system.Firstly,for the production of dataset,this project establishes the detection dataset and cow face recognition dataset for the training of cow face target detection model by collecting the cattle image data from the pastures in Chahar region of Inner Mongolia Autonomous Region,Sunit region of Xilin Gol league,and Tumut region of Baotou,respectively;secondly,the target detection model of YOLO series is used to realize the cow face target detection,and the cattle face target detection dataset is used to YOLOv5 is trained and analyzed to show that YOLOv5 performs well for bull face targets.Then the Inception structure is applied to cow face feature extraction,and the five networks,VGG-16,Resnet50,Inception_v3,Inception_Resnet_v1/v2,are trained using the cow face recognition dataset,and the comparison between the models is completed to select the model with the strongest cow face feature extraction ability.Through the cow face detection,the cow face facial picture is obtained,and the detected cow face is sent to the cow face feature extraction model to complete the cow face feature extraction,and it is easy to complete the cow face recognition and verification after getting the cow face features.In this paper,the obtained bull face feature vector and the bull face feature vector in the database are subjected to cosine distance calculation,and the corresponding threshold value is set to derive the final recognition or verification result.In this paper,we combine the bull face target detection model and bull face feature extraction model The bull face recognition system is designed,using Falsk and Vue frameworks to implement the front and back-end design respectively,and combined with MySQL database to provide bull building service,bull face verification service and bull face recognition service.However,the system can only be used on a small scale,which is not particularly convenient for operators,so Tensorflow Serving combined with Tornado is used to redesign and complete the cow face recognition system.The system deploys the model files to Ali cloud server for industrial level deployment.By testing the system in the natural farming environment,it shows that the system proposed in this paper can effectively complete individual recognition in the farming environment,which provides technical assurance for the later realization of healthy cattle breeding,precision breeding,detection of individual physical characteristics of cattle,and cattle insurance applications.