Research On Animal Facial Recognition Algorithm Based On Deep Learning

In recent years,with the rise of intelligent aquaculture and the continuous strengthening of food safety management in the country,accurate identification of animals has become an urgent problem in the industry.Currently,the main method of animal identification is to wear a radio frequency identification tag.This intrusive identification method is prone to cause discomfort to animals,the situation of biting and dropping of labels often occur.Face recognition is a kind of biometric technology.Migration is applied to animal identification to have significant advantages such as safety,convenience and non-intrusion.The traditional facial recognition method relies on manual extraction of facial image features,which requires a lot of field experience and unstable performance.As the most important technical means in the field of computer vision,deep learning can extract more essential features of facial images than traditional methods,and does not require manual participation.In this paper,based on the theory of deep learning,the research on pig facial recognition technology has been carried out and the following work has been carried out.(1)Training data collection.Currently,there is no public data set for animal facials in the field of facial recognition.In this paper,the facial data of two hundred pigs were collected on a commercial farm for the production of pig face data sets.The data set contains 50,000 images,including sows,nursery pigs and fat pigs.The distribution ratio of the three types of pigs is about the same.Face regions and feature points are labeled for the data set.(2)For the face detection and correction problem,a pig face detection and correction method based on improved multi-task cascade convolutional neural network is proposed.This method uses the three-level network to step through the face regions and key points in the image.Perform detection and screening,and use the positional relationship between facial key points for facial correction.The experimental results show that the improved network can accurately locate the pig face area and the face key points in the image,and use the affine transformation to correct the rotation of the pig face according to the positional relationship between the left and right eyes.(3)An improved model based on bilinear convolutional neural network was designed for the problem of fine granularity and small visual difference.The improved network uses two VGG-16 to extract different facial features of pig face.And extract the different features of the network at different levels.The experimental results show that the improved network can extract the features of the pig face with finer granularity and lock the positional relationship between different semantic features,and obtain better reeognition results for the pig face samples with different scales,poses and illumination.(4)The model was deployed on the embedded platform Jetson TX1?and the detection and recognition performance of the model was tested on the self-collected dataset and the crawled multiple scene datasets.The test results show that the model can identify the face of different illumination and posture offline at a faster speed,and the recognition accuracy reaches 95.73%.