Urinary Sediment Cell Classification Recognition System Based On SVM Algorithm Research

Urinary sediment system has appeared with people’s more and more highlyexpectations for medical health care. Software platform which combined with patternrecognition algorithm has realized automatic classifying urine cells and efficientdetection. Tangible components in urinary sediment types diverse, complex structure,complex fuzzy image background, easily to cause interference and miscalculation,increase the difficulty of the automatic identification of urinary sediment, and lowaccuracy. This paper researches of urine sediment recognition system from twoaspects of theory algorithm and actual clinical practice, using the adaptivetwo-dimensional entropy threshold segmentation and SVM classification algorithmtechnology, all kinds of visible part of the urine sediment cell image statisticalclassification.For Urinary sediment images processing, low power and high power usingcoordinate tracking technology to realize the perfect transition, using adaptivetwo-dimensional entropy Canny double threshold image segmentation, andmorphological characteristics at low magnification image Epithelial (Epithelial),Urinary cast (tube) automatically separate; By low power tracking regions to obtainthe most effective, the coordinates of the high magnification image with a fixedthreshold segmentation WBC and RBC of cell image extraction at high magnification26d characteristic vector, after normalization processing, using the SVM classifier toidentify the training and classification, the last successful classification, testing,recognition rate can reach more than90%.Urinary sediment cell defocusing of the image noise and the interference ofbackground, comparing several denoising algorithms, choose to use a Gabor filter ofimage preprocessing. In cell image segmentation, in performance and efficiency ofthe exploration, compare the segmentation experiments, and puts forward animproved algorithm to increase the speed of image segmentation and segmentationeffect. Urine sediment visible components are extracted on the basis of thecharacteristic value of use of the most representative of26D configuration, statistics and texture characteristics, improve the kernel function and parameters of the SVMalgorithm to obtain higher recognition rate of the SVM classifier.In this paper the entire urinary sediment image processing system, the validationof the SVM algorithm, and gradually improve the correct recognition rate of the cellsin the urine. Through the SVM algorithm under the condition of the training samplesis very low, getting the correct classification effect, which can promote to achievegood classification ability. In the normalized eigenvector matrix, using crossvalidation method to choose kernel functions and parameters, according to the Redcell (RBC), White cell (WBC) best26dimension feature vector to distinguish, wedesign a classifier, and get the corresponding confusion matrix. In the training of theclassifier to obtain the highest recognition classifier, and after the test, we verify thehigh accuracy of the classifier.