Research On Cow Identification Method Based On Neural Network

With the improvement of living standards,the demand for poultry and their supply of dairy products is increasing,and new requirements for modern healthy farming are put forward.The identification of individual cows in modern farming has become a research hotspot.The traditional classical methods have shown insufficient deficiencies in terms of input cost and actual use requirements.How to overcome the low recognition rate of dairy cows caused by the complexity of dairy cow image under the real-time requirement of the system has become the focus of this paper.The cow identification method based on convolutional neural network is a contactless cow identification method that meets the requirements of modernization and intellectualization.Thanks to the development of deep learning in image processing and pattern recognition.Based on the theory of convolutional neural network,this paper studies the problems of dairy cow individual recognition,and introduces a dairy cow individual recognition network suitable for complex background.The main research of this paper is as follows:1.The experimental images were collected from the farm of Mengniu Dairy Base in Hohhot,Inner Mongolia,from July to August 2018.The images of the trunk,head and tail of dairy cows were collected.A total of 170 adult dairy cows were collected,and 40 segments of 360 degree video data of dairy cows were captured.2.In this paper,a denoising method based on multi-scale convolution neural network(MSDCNN)is proposed.The image is convoluted and merged through convolution windows of different sizes.The features of the image are extracted from different perception domains.Relu is selected as the activation function of the network.The cow image is superimposed with Gaussian noise to simulate noise interference,and the denoising effect of MSDCNN is measured by denoising experiment.The peak signal-to-noise ratio(SNR)of MSDCNN for Gaussian noise denoising with standard deviation of 10,20,and 30 is 33.44,31.82,and 28.13,respectively.The Gaussian noise denoising peak SNR of traditional Gaussian denoising with standard deviation of 10,20,and 30 is 31.83,29.02,and 27.13,respectively.Therefore,MSDCNN shows a good denoising effect under the same data set.3.A cow-based individual recognition method based on deep learning and suitable for natural light images is used.In this paper,based on the complexity of the surrounding environment of cows' individual images,Google's Inception V3 model is used to abstract the cow images through multiple convolution layers,and finally connected to the softmax layer to achieve individual identification of cows.In this experiment,the video data of dairy cows were analyzed and processed to obtain21632 images of each individual cow.Then,the image data of the individual cows are randomly divided into 9:1,which are used as the training set and the test set respectively,wherein the test images of each cow individual are between 65 and 80 sheets.The recognition accuracy of each cow is counted.It is calculated that the recognition rate of the single-frame cow image reaches 87%,and there is no over-fitting in the network.Further,the multi-frame image recognition is based on Top-5 of Inception network probability data,and the individual recognition rate of cows reaches 92%.The accuracy of the algorithm was tested by cross-validation test,and ten sets of test accuracy data were given in this paper.This experiment shows that the deep learning-based method presented in this paper can accurately identify cows under complex backgrounds.4.The effect of noise on recognition accuracy is verified by adding different levels of noise to the original image.The statistical recognition rates of Gaussian noise images with superimposed standard deviations of 10,20,and 30 are 85.5%,84.6%,and 82.7%,respectively.The statistical recognition rates of images after denoising using MSDCNN algorithm are 86.4%,85.8%,and 84.6%,respectively.This proves that MSDCNN algorithm can effectively improve the recognition accuracy of cow noise image.The results of this study fully prove the feasibility of dairy cow individual recognition technology with complex background in video image,which provides an effective solution for dairy cow recognition and application basis for modern farming system.