| As a kind of animal that can be raised on a large scale,pig plays an extremely important role in the field of animal breeding.The rapid development of today’s society has significantly improved the daily living standard.The increase of pork supply has led to the continuous expansion of pig breeding bases.However,how to effectively and safely carry out intensive management of pig breeding bases has become the primary problem we are facing.Pork plays a very important role in the types of meat products consumed in China,but swine fever outbreaks occur from time to time.In order to ensure food safety and prevent problematic pork from entering the market,the need for effective supervision of pork source is increasingly urgent.In addition,with the continuous enhancement of people’s awareness of prevention,pig insurance is often carried out in the breeding base so that compensation can be made after the accidental death of the pig,so as to further reduce the risk of breeding.However,with the increasing number of pigs in the breeding base,traditional breeding methods often lead to insurance fraud and other problems.How to effectively manage pigs is of great significance to both the breeding base and the insurance company.In addition,with the continuous increase in the size of the base and the surge in the number of pigs,seeking effective pig management measures is an effective way to reduce the cost of breeding and improve the quality of breeding.To sum up,intelligent management of pigs is the key means to solve the above problems,and pig identification is the basis of intelligent management.Pig face recognition is a reliable technology to solve the problem of pig identity recognition,and the maturity of face recognition technology provides the possibility of pig face recognition,among many face recognition methods,the method based on deep learning has been widely used.In addition,convolutional Neural Network(CNN)has gradually become one of the most important method in deep learning,and various excellent models have excelled in all kinds of authoritative international competitions.Pig face recognition is carried out on the basis of pig classification,so this paper carried out relevant research on pig face recognition method based on convolutional neural network,and the main work contents are as follows:1)There is no publicly available data set of pig faces.The data set in this paper is obtained by means of camera timing shooting.Direct contact with pigs is avoided.The whole period of shooting is one and a half months.In addition,according to the living habits of pigs,photographing is conducted during their active period,i.e.the feeding stage,which not only reduces the space occupied by the hard disk,but also ensures the maximum diversity of facial postures of each pig with Identity Document(ID).2)The quality of the data set is critical to identifying the performance impact of the model.In order to improve the data quality and enhance the performance of CNN recognition model,a series of processing is carried out on the initial data set in this paper.In this paper,background information removal and image reduplication are carried out on the data set.By removing interference factors in this way,the CNN model can effectively learn the facial features of each ID,and prevent model overfitting through image reduplication to ensure the robustness of the CNN model.In addition,in order to ensure the maximum facial feature information of each ID pig when the images enter CNN for supervised learning,data analysis is conducted on the size of all images in the training set.The most reasonable normalized size is established according to the size distribution of the data,and the equivalent constant pixel value filling method introduced in this paper is used to normalize the size of the image to ensure that facial features do not change due to the size normalization.3)During the filming process,some pigs died,were moved out,and new pigs were purchased.The above factors cannot be avoided,and will also occur in the process of large-scale breeding,resulting in the uneven distribution of facial data of each ID pig.In order to avoid the problem of poor performance of the recognition model caused by data imbalance,in this paper,a multi-classification weighted loss pig identification method based on skewness analysis is proposed.In this paper,the SE＿RESNext50 network was used to extract the facial features of each ID pig.In addition,according to the data collection of each ID pig,improve the network’s attention to categories which below skewness analysis values,and restraint weights higher than the skewness analysis value categories.Finally,the identification model can accurately extract the facial features of each ID pig.For 63 reproductive sows,the precision of the proposed pig facial recognition method can reach 96.69%,and the comparison with other common methods proves the superiority of the proposed method. |
