Research On Formation Detection Algorithm For Urine Sediment Image Based On Machine Learning

The formation of various components in urine sediment has important diagnostic and differential effects on human kidney and urinary system diseases.Red blood cells,white blood cells,epithelial cells and casts are often detected in clinical practice.The tube types are divided into transparent tube type and granular tube type.Types,the number of dif-ferent types of casts have important clinical diagnostic reference significance for urinary system diseases.The image of urine sediment has many kinds of forming points,the texture contour is ever-changing,the image clarity is not uniform,the color brightness distribution is uneven and the defocus is serious.The above characteristics lead to dif-ficulty in identifying various types of formation in urine sediment and slow recognition speed.In view of the above characteristics,this thesis firstly studied the morphological changes of urine sediment image,and processed the image by image enhancement,edge detection and segmentation,and then extracted the feature matrix of the image.Finally,it was classified by machine learning-based combination algorithm.The tube type is clas-sified into a transparent tube type and a particle tube type in detail.The main research contents are as follows:Urinary sediment image pretreatment and initial screening.The urine sediment im-age obtained by electron microscopy has problems such as serious noise,strong impurity interference,low contrast,and uneven illumination.Firstly,the original image is denoised,enhanced and cell segmentated,so as to complete the initial screening of all urine sediment image formation,and the various forms of urine sediment are formed.The processing in-volves filtering algorithm and binary value.The algorithm,edge detection algorithm and morphological processing methods are used to detect the different urine sediments under high and low magnification.The Canny operator and the watershed algorithm are com-bined to detect the edge.The urinary sediment after image processing has feature extraction of the formed fraction.After the original image is segmented into a single object image,the geomet-ric,texture,and statistical feature matrices of the image are extracted.The feature matrix data is sent to the classifier for learning.According to the learning results,the rationality of these features and the influence of various feature dimension combinations on the fi-nal classification effect are studied.Finally,the 41-dimensional features with the highest recognition accuracy are selected.In addition,in view of the large amount of Gabor fea-ture data,it is proposed to adopt the(2D)~2PCA dimension reduction method to improve the efficiency of the algorithm.According to the characteristics of different formation under high and low magnifi-cation,a combination algorithm based on morphology,decision tree and SVM classifier is proposed.The cross-validation method was used to obtain the optimal parameters of the classifier,and the red blood cells,white blood cells,epithelial cells and casts in the urine sediment image were effectively classified,and the tube type sub-class transparent tube type and particle tube type were more detailed.Identification and classification,the overall classification recognition rate reached 95%.