The Research Of The Urinary Sediment Images Automatic Recognition Algorithm

The automatic urinary sediment analyzer is an effective method for automatic urinary sediment examination, this paper starts to study the research of urinary sediment images automatic recognition algorithm. During the project ,the author took over lots of the relevant good achievements, according to the characteristics of the urinary sediment images, studied the suitable and feasible method to enhance the performance of recognition in many aspects of image recognition such as image enhancement, image segmentation, feature extraction and selection, image recognition. Based on the study, the author proposed the new urinary sediment images automatic recognition algorithm based on combinational image segmentation algorithm, modified feature extraction and feature selection algorithm and BP neural network classifier. Besides, the author tested the algorithm using lots of urinary sediment images based on MATLAB. The results showed the algorithm can recognize the four kinds of elements and impurity in the images, can attain high recognition accuracy for the four kinds of elements, especially for the casts and epithelia.During the research, the author proved the feasibility and effectiveness of the algorithm, finished the project. The major contribution of the dissertation is that: 1) proposed urinary sediment images double combinational segmentation algorithm. The algorithm adopted wavelet transforms plus morphology processing to do coarse segmentation for urinary sediment images, reduce the effect of defocusing; according the subimage from coarse segmentation, adaptively use the modified adaptive thresholding plus morphology processing or canny detection plus morphology processing to do fine segmentation to get satisfying segmentation result; the algorithm used'peel off'algorithm to deal with the overlapping particles, thereby improving the application of the algorithm. 2) proposed urinary sediment image double adaptive thresholding segmentation algorithm. The algorithm adopted wavelet transforms plus morphology processing to do coarse segmentation for urinary sediment images, reduce the effect of defocusing; according to the subimages attained from coarse segmentation, acquired the corresponding subimages after wavelet transform, then did adaptive thresholding plus morphology processing to get segmentation result. The algorithm has the fast speed advantage, and can get satisfying segmentation result for not too complex urinary sediment images. The'peel off'algorithm can be used here for dealing with the overlapping cells segmentation. 3) according to feature selection issue, proposed new feature selection method based on modified genetic algorithm. The algorithm adopted hierarchical classification idea, simplifying the urinary sediment feature selection problem; put the bi-directional method and simple genetic algorithm together, realizing gene gradual fixing technology and improving the performance of genetic algorithm; introduced niche technology, keeping the diversity of population, preventing from falling into local extremum; introduced feature space partition method, optimizing the initial population; adopted adaptive mutation operator, improving the local searching capability of genetic algorithm; introduced multi-criteria , improving the stability and classification ability of the feature subset attained.The structure of this dissertation is as follows: chapter one illustrates the meaning and the objective of the project, and relevant the current research situation; chapter two illustrates the research of urinary sediment image enhancement algorithm, including relevant theoretical analysis, experimental result and the realization of the algorithm; chapter three illustrates the research of urinary sediment image segmentation algorithm, including relevant theoretical analysis, experimental result and the realization of the algorithm; chapter four illustrates the research of urinary sediment image feature extraction algorithm, including relevant theoretical analysis, experimental result and the realization of the algorithm; chapter five illustrates the research of urinary sediment image feature selection algorithm, including relevant theoretical analysis, experimental result and the realization of the algorithm; chapter six illustrates the research of urinary sediment image BP neural network classifier, including relevant theoretical analysis, classifier design and experimental result; chapter seven illustrates the major contribution of the dissertation and the future work.