Application Of Support Vector Machine And Fuzzy Theory For Remote Sensing Image Classification

Remote sensing image classification is a key technology in remote sensing applications . Rapid and high-accuracy remote sensing image classification algorithm is the precondition of kinds of practical applications . Traditional pattern classification methods are based on the principle of Experiential Risk Minimization , they can achieve the best result , only when the number of samples approaches infinity . Unfortunately , the number of samples is actually limited and the data dimension is high . Taking into account the good generalization of support vector machines in small samples , nonlinearity and high dimension space , and according to features of remote sensing image , this dissertation deeply studies support vector machines and their application for remote sensing image classification . The main contributions of this thesis are given as follows :(1)We discuss several training algorithms of SVM , and study the QP problem of SVM, and then select the SMO method as the QP algorithm .(2)We deeply study multiclass algorithm of SVM . The application of two classifiers for remote sensing image classification , namely one-against-one SVM multiclass classifier and one-against-the-rest SVM multiclass classifier , is introduced in detail . Then aiming at the case that existing the missed samples or the multivocal samples via two kind of multiclass SVM classifier , using fuzzy method to reclassify these samples . Experiment result suggest the mixed classifier may improve the accuracy of remote sensing image classification .(3)We discuss two kind of theory of Fuzzy Support Vector Machines(FSVM) and improve training algorithm of one FSVM which import fuzzy gene in training process , and then put forward edge-effect training algorithm for FSVM .(4)Some useful study aiming at decision tree SVM multiclass classifier is bring to success . According to experiment result via iris dataset and remote sensing image dataset, we give an effective class-distance method to select a better hierarchical structure of decision tree .