Research On Facial Feature Recognition And Segmentation Algorithm Of Horses Based On Convolutional Neural Network

In recent years,as the country and the XinJiang Uygur Autonomous Region attach great importance to the development of the horse industry,in order to ensure the continuous development of the horse industry in Xinjiang,it is urgent to implement the registration of horse breeds in XinJiang.At present,the registration of domestic horse breeds mainly depends on the traditional manual registration method.The registration agency will register and store the characteristic data of horses in conformity with standards in the paper version of the registration book.In order to realize the electronic registration of horse breeds and information sharing,it is necessary to research and develop an electronic registration system for horse breeds to register and manage horse breeds.passport.However,at present,the appearance feature extraction methods in horse breed registration and passport information are mainly done manually,which requires text description and tracing marks on the horse’s facial features.There are problems such as large workload,low efficiency,low accuracy,and high labor costs.In order to solve the problem of extracting the facial features of horses in the horse breed registration system and passport generation and issuance.This paper studies the extraction,segmentation,preservation of horse facial features and the output of passports in the horse registration system.The main contents and results are as follows:(1)The Mask R-CNN instance segmentation model based on convolutional neural network is used to identify and segment the horse face and horse facial features.First,obtain frontal images of horse faces through web crawlers and horse farm field shooting,use Labelme to create data set labels,and extract the features of the Region of Interest(ROI)from different levels of the feature pyramid through the backbone network of ResNet-50-FPN,and finally The bounding box recognition and mask prediction applied to each ROI separately through a fully convolutional network to generate the corresponding horse face and horse face feature mask to achieve the segmentation of the horse face and horse face features in the image and the background.(2)Constructed a data set of horse face and horse face facial features with segmentation annotation information,used to train the corresponding model.The average precision(Average Precision,AP)of the target detection evaluation index of the COCO data set is used to test and evaluate the model in this paper on the test data set.Among them,the horse face segmentation evaluation index AP reaches 86.006%,the star reaches 44.962%,and the star reaches 22.929%,Slender meteor nose white reaches 39.475%,long meteor nose white 30.463%,Changguang meteor nose white 28.446%,white face 29.221%.Experimental results show that the method in this paper has a better horse face detection effect,and can achieve pixel-level horse face and horse face feature information segmentation at the same time with more accurate detection.(3)Completed the development of the horse breed registration system using the Django web framework,and deployed the horse characteristic identification and segmentation model based on the convolutional neural network training in the horse breed registration system.The result of dividing the image is stored in the horse breed registration system,which is used to generate the electronic version of the PDF horse passport.Compared with the traditional manual marking method of facial features of horses,this paper studies the horse’s distinctive feature algorithm and combines it with the horse registration system for practical application,which shortens the working time period of marking the appearance of horses.Up to now,according to the background data statistics of the system,there are 4652 pieces of horse information registered in the horse breed registration system,and 135 electronic passports of horses have been generated.