Research On Sheep Face Recognition Based On Key Point Detection And Euclidean Space Measurement

The Inner Mongolia Autonomous Region is a important farming and aquaculture areas in our country.Sheep production is the landmark industry.With the development of information technology.Digitization,intelligent and automation become the main direction of modern aquaculture.At present,the PIV(personal identity verification)of sheep mainly uses the method of radio frequency identification.But the method has many disadvantages,such as high cost,damage to sheep,drooping easily and substitute.Therefore,based on the unique biological characteristics of sheep face,the research of non-contact sheep individual recognition using computer vision has made great progress.This paper uses the advantage of deep learning in graphic image processing technology to study the way to sheep face recognition which is based on key point detection and euclidean space metric.Because there is no huge amounts of data.The way to get primary data is self-collected data.Then,sift out the available data,expand the data and finish building the sheep face data sets.In order to solve the problem of redundant information interference in the image,a processing method of first detection,then alignment and finally recognition is proposed.We use the target detection algorithm to extract the sheep face region in the image,then use the key point detection algorithm to locate the key points of the face,then use the face alignment algorithm to align,and finally use the aligned sheep face to complete the recognition.The self-made sheep face detection data set is used to train the sheep face detection neural network model based on more anchor frame number,different coverage threshold and different detection frame merging algorithm;The self-made sheep face key point data set is used to train the sheep face key point detection algorithm;The self-made sheep face recognition data set is used to train the sheep face recognition neural network model based on Inception-V2 feature extraction network and center loss function optimization to solve the European space measurement problem.Through the above combination scheme to achieve the best sheep face recognition effect.In this paper,the sheep face target detection model,the sheep face key point detection algorithm and the individual identity recognition algorithm are combined,and the self-made data set is used for training.Finally,the open set sheep face recognition is realized by measuring the Euclidean space distance of different sheep faces.The experimental results show that the method can effectively identify individuals in the actual scene,and has good engineering application value.