Research On Recognition Of Milk Somatic Cells Based On Binary Model

Mastitis is a multiple disease that causes yield reduction in the process of cow breeding.In medicine,the number of macrophages,neutrophils,lymphocytes and epithelial cells in milk somatic cells are analyzed to judge the health status of dairy cows.In order to provide a new diagnostic technique of dairy cow mastitis and improve the classification efficiency of bovine milk somatic cells,the following works were carried out by using image processing technology.(1)A variety of preprocessing methods were used to process the milk somatic cell image to enhance the image display effect and reduce the difficulty of obtaining the key information of the image.(2)The milk somatic cells recognition algorithm based on CNN is proposed.CNN has the ability to extract deep-seated information of images,and abandons the process of manual feature extraction in traditional algorithms,so it gradually becomes the research focus in the field of image processing and recognition.In this dissertation,a shallow CNN network based on the characteristics of milk somatic cells image is constructed,and the extended image database obtained by image enhancement is used to complete the model training.Through experimental comparison,the convolution neural network model can effectively identify the types of bovine milk somatic cells,which provides a new idea for the follow-up classification research.(3)The milk somatic cells recognition algorithm based on texture feature fusion of LVF and RF dimensionality reduction is proposed.Aiming at the limitation of HOG and LBP,LVF and RF are used to reduce and fuse the texture features extracted by HOG and LBP respectively,and then BPNN is used for classification.The experimental results show that the dimensionality reduction fusion method proposed in this dissertation is feasible,and the texture features are suitable for the recognition and classification of milk somatic cells.(4)In this dissertation,a milk somatic cell recognition algorithm based on dichotomy model is proposed,and a three-layer dichotomy model of tree structure is constructed.Firstly,the morphological features of four kinds of cells are extracted,and three kinds of features are selected by RFFS,and the first two layers of the model are constructed with BPNN.Then,according to the fact that it is difficult for macrophages and epithelial cells to classify through morphological features,the third layer of the model is constructed by texture features and shallow CNN network respectively.The experimental results show that the recognition effect of the dichotomy model of tree structure is ideal.