Abstract: Aiming at the problem of wide demand of  livestock facial recognition  in the breeding industry, we  proposed a  feature point detection method of pig face based on convolutional neural network, which solved the problem that it was difficult to detect feature points of pig face. Firstly, the pig face data was collected and the feature points were marked, and a new collection method was used to solve the problem that the pig mouth was usually invisible. Secondly, we calculated the structures of the pig face data and the human face data,  matched the pig face and human face with high similarity, and constructed the pig face and human face matching data set. Thirdly,  TPS (thin plate spline) deformed convolutional neural network was trained by matching data set,  and  the deformed pig face data set was obtained to fit the  feature point detection model of human face. Finally, the  feature point detection neural network model of human face was  fine-tuned by using the deformed pig face data set, and  feature point detection model of the pig face was obtained. The experimental results show that the error rate is only 5.60% by using the proposed method to defect feature points of pig face.

Key words: feature point detection of pig face, convolutional neural network, feature point data set of pig face, deep learning