Research And Implementation Of Pig Face Individual Recognition Method Based On Hierarchical Convolutional Neural Network

In recent years,animal husbandry is developing in full swing.Pork is widely loved by Chinese people because of its rich nutritional value.Because of this,pig breeding,which is an important part of animal husbandry,has also received extensive attention.With the expansion of pig breeding scale,problems such as pork food safety and breeding insurance fraud have emerged.The current priority is to adjust the pig breeding model scientifically and effectively.The rapid development of computer vision and deep learning technology has provided more possibilities for intelligent farming.The intelligent breeding of pigs is the product of the effective combination of technology and traditional animal husbandry.It can track the whole process of pig breeding,and choose different breeding programs for different pigs.While saving labor costs,it is more scientific and effective to carry out intensive management of pig breeding.The realization of pig individual identification is the premise of intelligent pig breeding,so how to use deep learning and computer vision technology to carry out individual identification of pigs has become a hot research topic at present.For pig individual identification,the pig’s face is the area with the greatest difference in characteristics relative to other positions,so this paper uses the pig’s face to carry out the task of pig individual identification.In this paper,based on the method of hierarchical convolutional neural network to identify pig faces and determine pig IDs,the following work is carried out:（1）At present,there is no public dataset of pig faces,so the data set studied in this paper is collected and made by ourselves.The collection of the dataset took 3months in total.The collection method was manual shooting,and taking photos of pig faces from multiple angles to enrich the diversity of the data.Due to the complex real environment,pigs cannot cooperate like humans,and many conditions are out of control,so there are some pictures that do not contain pigs’ facial contours in the dataset.In this paper,a SE-Res Ne Xt network model is used to implement a binary classifier to filter the data.Aiming at the problem of uneven number of pictures of different pigs,data enhancement technology is applied to balance.Finally,the processed image and the corresponding pig ID number are combined into a dataset file,which is convenient for model training.（2）The living conditions of pigs are complex under real conditions,and the background information is not conducive to the individual identification of pigs.In this paper,a hierarchical convolutional neural network model is proposed for the task of pig face recognition in complex real environments.The model is composed of two levels of pig face location detection network and pig face recognition network.The pig face location detection network is the Yolo V3 network,and the pig face recognition network is the SE-Xception network improved with multi-scale feature pre-extraction in this paper.The hierarchical convolutional neural network model was used to test on the dataset collected in this paper,and the recognition rate reached93.83%,And compared with other neural networks with superior performance,the performance of the model is evaluated,which proves the superiority of the method in this paper.The hierarchical convolutional neural network proposed in this paper has achieved a good recognition rate in the pig face recognition task,and can perform the pig face recognition task under realistic and complex conditions.It is a meaningful exploration of deep learning algorithms in livestock face recognition work.