Research On Cattle Individual Identification Based On Computer Vision

In today’s society,science and technology are developing rapidly,and computer technology and Internet technology are constantly penetrating into all areas of the national economy,and are closely connected to the world.At the same time,the "Internet +" model has had a huge impact on people’s production and lifestyle,and has accelerated the transformation of traditional industries,especially in the field of animal husbandry.Traditional animal husbandry is developing towards intelligence,precision,and scale.The identification of individual livestock is the core content of precise animal husbandry.This paper proposes a method of identifying individual cattle based on computer vision based on non-contact image recognition.The method for the individual identification of cattle in small and medium-sized pasture environments shows great advantages and is worth popularizing.The research scheme of this paper mainly includes three parts: the establishment of data sets,the design of target detection models based on YOLO v3,and the design of individual identification models based on dictionary learning.The first is the establishment of data sets.In this project,we collected field image data of cattle in the Chahar and Sunite areas of the Inner Mongolia Autonomous Region,and respectively established a cattle face detection data set for the detection model and a cattle face image dataset for the identification model.Secondly,using deep learning technology combined with target detection theory to detect cattle face parts.By comparing the detection results of different models,we choose to improve on the basis of the YOLO v3 detection network.The target detection model is trained on the established cattle face detection data set,and the experimental results are summarized through analysis.Then,the dictionary learning theory is applied to the individual identification of cattle,and an enhanced dictionary learning algorithm(CTDLCKSVD)based on Cartoon-Texture Decomposition and Label Consistent K-SVD is proposed.Based on the LC-KSVD algorithm and the Cartoon-Texture Decomposition theory,the cartoon component and texture component are extracted to obtain the structural and detailed information in the image,and the input matrix and the augmented learning dictionary are reconstructed.Under the premise of ensuring the recognition speed,the accuracy of cattle face image recognition is effectively improved.In this paper,a cattle face target detection model and an individual identification model are combined,and an individual identification system for livestock is built and tested in the actual natural breeding environment of the pasture.The results show that this method can effectively identify individuals in the pasture environment.This has an essential impact on the precision breeding,monitoring of livestock signs,management of traceability data,and prevention of fraud claims.