Research On Pig Face Recognition Based On Convolutional Neural Network

The 21st century is the era of artificial intelligence.With the development of artificial intelligence technology,the rise of intelligent breeding industry and the continuous strengthening of national food safety control measures,accurate identification of animals has become an urgent task for the industry.Currently,the main method of identifying livestock is to punch and paste the wireless identification mark.This kind of cumbersome identification method is easy to cause discomfort to livestock,and the label often falls off.Face recognition is a biometrics technology that has good mobility when applied to livestock identification.Its advantages include security,convenience and non-invasive.Traditional face recognition methods rely on the manual extraction of face image features,which requires a long time of experience accumulation,and the effect is not obvious,unstable.As the most important technical tool in face recognition field,convolutional neural network can automatically extract more and more complex features of images,thus greatly improving the accuracy of the algorithm.Based on the theory of convolutional neural network,this paper studied the technology of pig face recognition^[1].As for data collection,there is currently no public data set of pig face in the field of animal recognition.In this paper,facial data of 30 pigs used to make pig faces were collected in the pig farm.This data set contains 30,000 images,and regions and feature points of pig faces were marked on the data set during the whole experiment.To solve the problem of pig face detection,MTCNN^[2]face detection method（the starting point of the network design is face detection）is used,and this technology is applied to the pig face detection as a transfer technology.This method uses multi-stage network to gradually detect the pig face area and key points in the image.Filter the position relation between the important points of the face and use it to correct the face.Through training experiments,MTCNN can accurately identify the pig face area and key points in the image.An improved model based on deep convolutional residual neural network is proposed to achieve good pig face recognition.The model uses an improved global convolution residual network to extract more and more complex facial features.Experimental results show that the improved network can extract better features of pig face.The model was deployed on the cloud platform,and the detection effect and recognition performance of the model were tested on the data set.The test results show that the model can lock the position of the face well and recognize the pig’s face effectively,with the recognition accuracy reaching 92.3%.