Research On Cattle Individual Recognition Technology Based On Cow Nose Texture

At present,China's animal husbandry presents a large-scale and intensive development trend,and carrying out individual identification with the manual way has been unable to meet the demand.Therefore,the individual identification system is required for modern management.In recent years,the cattle individual identification system based on Radio Frequency Identification(RFID)technology has been applied in some cattle farms.However,RFID tags are not only easy to be damaged and lost,but also face the risk of information tampering.To solve this problem,in this thesis,a cattle individual identification method based on muzzle images is proposed.Muzzle pattern is a kind of biological feature with uniqueness,which is difficult to forge and modify.In addition,the image recognition system has low cost and high speed,and it can obtain remarkable results.The main work of this thesis is as follows:First of all,this thesis proposes a cattle individual identification method based on feature fusion,which adopts local binary pattern and Weber local descriptor to extract texture features of muzzle,respectively.Due to the dimensions of the extracted features are relatively large,the principal component analysis algorithm is adopted to carry out dimension reduction of the two kinds of features.Finally,the support vector machine is applied to cross-validate the classification of muzzle images.In this thesis,the gradient direction calculation of Weber local descriptor method is improved,so as to improve the recognition effect.The experimental results indicate that the accuracy of the cattle individual identification method based on feature fusion is 98.4%,which is higher than that of the cattle individual identification method based on single feature,and this method still has high robustness in the case of image rotation.Secondly,in the cattle individual identification method based on convolutional neural network,this thesis designs a CNN-12 convolutional neural network model,which adds a batch normalization layer after the convolutional layer to accelerate the training speed,and adds a Dropout layer after the full connection layer to avoid overfitting.In addition,the influence of different pooling function,activation function,optimizer and batch-size value on the accuracy is verified through experiments.Finally,the accuracy on the self-built muzzle image data set reaches 98.99%.Thirdly,this thesis proposes a cattle individual identification method based on transfer learning,which applies the idea of transfer learning to the muzzle image recognition for the first time.The experiment of transfer learning is carried out by using multiple models,and the model is adjusted and optimized.The experiment proves that the method of transfer learning has a good effect,higher accuracy as well as less training time.Finally,the MobileNet model is used in the transfer learning method,with an accuracy of 99.26%.The model has a small size and can be deployed to the mobile phone terminal,so it is convenient for practical application.Fourthly,as muzzle image has no public data set,62 kinds of muzzle image data set is built by ourselves.In the cattle individual identification method based on feature fusion,gray processing is carried out to the image,and then the contrast limited adaptive histogram equalization algorithm to enhance the image contrast and make the image clearer.In the method based on the CNN-12 convolutional neural network model and the method based on transfer learning,the image is normalized in size and the data set is expanded with the data enhancement method.