Feature extraction and lightweight methods for cattle facial recognition

CSF300 is a standard dataset for cattle facial recognition scenarios, which includes 300 individual cattle with a total of 2,899 facial images, covering individual cattle in cattle breeding facilities.

Therefore, to verify the generalization of the cattle facial recognition model, this project introduces the BCFD-10K and CFFD datasets as supplements.

The design of the hybrid enhancement module MEM is the core of the cattle facial recognition network DSFFNet.

The use of focus loss function can effectively adjust the sample weights for the imbalanced number of cattle sample images in some cattle facial recognition datasets.

Therefore, in a limited number of training epochs, the cattle facial recognition model can achieve optimal recognition accuracy.