Few-shot Classification with Shared and Private Branches for Cow Face Recognition

TO reduce the significant losses in cattle farming, the government has carried out necessary policies, the most important and effective way is to help farmers buy insurance for their cattle. Thus, it requires advanced technology for cattle identification. Among various methods of cattle identification, artificial intelligence (Al) technology yields fewer negative effects and lower costs when compared to other approaches. However, since the gathered cattle pictures are limited, directly utilizing state-of-the-art Al techniques for cattle identification often leads to unsatisfied results, and is prone to overfitting. In this paper, we propose a novel cow face recognition method, helping to build up an intelligent livestock platform. Based on the limited cow pictures, our model extracts two elements, i.e. the shared and private features, that are independent of each other. This makes it possible for the model to learn from a small amount of samples and classify images more accurately. Our model outperforms the traditional few-shot classification methods depending on our cow face dataset; and can identify cattle more accurately.