A Deep Detection and Identification Framework for Smart Sheep Farm

In the field of precision livestock farming, sheep management usually employs ear tag. However,ear tagging leads to a number of problems, such as high cost, time consuming, tag loss. Therefore,automatic recognition of individual sheep becomes an urgent problem to be solved. The realscenes include the following characteristics, nonuniform illumination, complex poses, occlusionsand so on. And these characteristics bring about great challenges to sheep face recognition. Inthis paper, we collect sheep face dataset automatically via designing and coding a procedure,then recognition task is carried out via constructing a robust sheep face recognition model. Inparticular, You Only Look Once Version 5 (Yolov5) model is improved via employing the attentionmechanism, by doing so, the improved YOLOv5 model is further enhanced via equipping withContent-Aware Reassembly of Features (CARAFE) operator. Furthermore, the proposed modelis trained and verified to optimize the involved parameters and improve the robustness of themodel. Based on the proposed framework, the sheep face can be recognized efficiently, and theaccuracy is up to 99.39%, which is better than the state-of-the-art methods. The research resultsare helpful for precision breeding.