Open Set Sheep Face Recognition Based on Euclidean Space Metric

Figure 6

SheepFaceNet network structure: deep architecture refers to feature extraction networks, such as inception, ZFNet; L2 refers to feature normalization (make the length of its eigenvector 1, so that all feature points will be mapped to Euclidean space vector); embedding refers to the feature vector after L2 normalization, and this feature vector represents a sheep face picture; center loss refers to the center loss function [32], which specifies a category center for each class (in the sheep face recognition model, one class corresponds to a sheep). The corresponding features of the same type of image should be as close as possible to their category center, and the category centers of different categories should be as far away as possible. The center loss can make the training data have “cohesion.”