The Study Of SVM-based Recognition Of Particles In Urine Sediment

Currently with the convergence of the medicine and the computer science, medical image based computer aided diagnosis (MIBCAD) has developed rapidly and become a new hot research area. Computer aided diagnosis (CAD) can help doctors to identify or classify the medical images quickly and efficiently, also can improve the accuracy.After the pretreatment and features extraction, it is preferred to use methods of pattern recognition for classification and recognition. Classification methods commonly used before are based on traditional statistical theory or based on infinite training samples, such as ANN. But in practice application, the number of the sample is limited. Therefore, the traditional methods are inclined to bring many problems like model-choosing, over-fitting, non-linear, disaster of dimensionality, local minimum.Support Vector Machine (SVM) is proposed by Vapnik in 1992-1995. SVM is based on Vapnik-Chervonenkis Dimension theory and Structural Risk Minimization (SRM) principle in Statistical Learning Theory (SLT), and SVM focuses on the optimal solution of limited training samples. The improvement of SVM has successfully solved the puzzles in many other machine learning methods. So with the development of theory and research, SVM becomes more and more valued.In this thesis, SVM as a new pattern recognition method is brought into recognition and classification of particles in urine sediment. First, we introduced the theory basis of SVM, emphasized the key points of Statistical Learning Theory (SLT), the mathematic model and kernel function principle of SVM. Then the progresses of SVM research are proposed, and several improved SVM training arithmetic and the attributes of them are analyzed. At last, after feature extraction, we use 1-v-1 strategy and improved SMO arithmetic with LIBSVM to solve the multi-class problem. Cross-validation method and the contour chart of the accuracy have been implemented to select the kernel function and the parameters of SVM. A method of two-level-classifier is implemented, and the obtained results provide high classification performance. Moreover, we bring the Classification-matrix to analyze classification accuracy of every class of the particles in urine sediment. And the performances of the BP classifier and SVM classifier have been evaluated. In this study, improved SVM method is brought into recognition and classification of particles in urine sediment and every experimental result has demonstrated that SVM classifier is quite promising and more effective.