Cattle Face Recognition Method Based on Parameter Transfer and Deep Learning
Hongyu Wang, Junping Qin, Qinqin Hou, Shaofei Gong
Abstract
Abstract In order to accurately identify cattle, help insurance companies to deremine claim of Cattle Insurance, or realize inteligent breeding. We need a cattle face recognition technology. Due to the difficulty in data collection of cattle face and the small amount of data, it is difficult to apply deep learning method to cattle face recognition. So there is a method combined with tarnsfer learning effectively trains the model,or initializes the network weights by parametric transfer. This paper proposed using VGGFace dataset pre-training network to solve the problem of small sample face recognition. The VGG-16 deep convolutional neural network model was used to extract the cattle’s face features. The model parameters obtained by training the VGGFace face dataset were used to initialize the network weights, and then the network was trained on the cattle face dataset. The experimental results show that the method can obtain 93% recognition accuracy under small sample data.