Reaserch On Facial Recognition Of Albasian Velvet Goats Based On Deep Learning

Under the background of digitalization of livestock market,individual livestock behavior monitoring,disease prevention and precise management become urgent problems for developing smart ranch,and individual identification becomes the prerequisite for solving the above tasks.Owing to labor-intensive and low efficiency,the traditional livestock identification method is insufficient the requirements of modern ranch development.Therefore,this paper proposes a deep learning-based facial image recognition method for Albasian velvet goats after analyzing and studying animal recognition and face recognition.The main work of the paper is as follows.Firstly,a sheep face target detection dataset is constructed for the sheep face detection problem,and a MRFEM-YOLOv4-based sheep face detection algorithm is proposed.Through designing a MRFEM module to YOLOv4 model,the proposed MRFEM-YOLOv4 can efficiently aggregate the feature mapping between different layers of the neural network.In addition,optimization strategies such as prior frame clustering are adopted to improve the sheep face detection performance,resulting the detection accuracy of the proposed MRFEM-YOLOv4 reaches 97.11%.Secondly,the sheep face key point detection dataset is constructed for the sheep face alignment task.Due to the weak differentiation of the local information in the sheep face key points,it is difficulty to detecting different key points.Therefore,this paper applies the Open Pose model to realize the 5-point localization of the sheep face key points.Moreover,the two-eye center linkage-based algorithm is employed to realize the sheep face alignment after extracting the key points.The sheep face alignment lays a solid foundation for the sheep face recognition task.Finally,the sheep face recognition dataset is constructed for the open set pattern recognition problem of sheep face recognition,and an ARM module is proposed to enhance the feature extraction ability of the backbone network.Furthermore,the Circle loss metric learning loss is applied to the sheep face recognition task for the first time.Comparison experiments show that the ARM＿Res Net50+Circle loss sheep face recognition model obtains the best recognition result,and the accuracy of sheep face recognition reaches 96.12%.