Region Multi-center Classification Method For Land Use Of Remote Sensing Images Based On Fuzzy Rough Set

With the rapid development of remote sensing technology, landuse classification is an important area of remote-sensing application. How to improve the landuse classification accuracy of remote sensing imagery is the key issues to be resolved currently.Rough sets and fuzzy sets are the mathematical analysis theory dealing with unceriain description of objects. Rough sets taking knowledge as a tool for object classification can directly mine knowledge from the given data without any preliminary or addtiional information,so it is very objective. However,it is necessary to discretize the continuous data before the knowledge reduction with rough set, which will cause a certain degree of information loss. fuzzy sets focus on the ambiguity of set, representing an imprecise concept by the notion of a membership functions. The combination of fuzzy sets and rough sets can solve the problem of information loss in the discretization process through transforming the data into the corresponding fuzzy membership by fuzzification process.This paper at first analyzes the spectral characteristics of multispectral remote sensing imagery,clusters samples with a certain distance threshold, calculats the distance distribution between the intra-class center vectors and the LU classificatory center vector, studies the relationship between the feature vector and the LU classificatory, confirms the LU classificatory pattern,and then extracts the region multi-center feature taking pixel region as the classification cell, builds the classification rules with the type amounts of intra-class center and the percentage of the pixels belonged to the class from the whole region pixels,and finally classifies LU categories by the classification rules.Region multi-center(RMC) method expresses sufficiently the basic idea of LU classification system and classification based on the area ratio of ground objects to whole region. In the meanwhile,this methed can effectively solve the problem that a convenient multivariate statistical model is in general not available for the multi-spectral feature of LU class beaucase that one LU class is made up of several ground objects. But this method is lack of sufficient theoretical support, and greatly increases the workload and the complexity of classification beaucase that the acquisition of correct classification rule is based on many times of experiments.Therefore,based on the Region multi-center (RMC)method,the paper fuses the fuzzy rough set theory, establishs classification rules based on fuzzy rough set model, does research on LU classification of the surrounding area in Changsha by using TM remote sensing image. This paper at first gets classification attribute data by Region multi-center(RMC) method, fuzzifies attributes according to the fuzzy set theory, establishes LU classification decision table on this basis.Then decision table attributes are reducted by a quickly attribute reduction algorithm based on rough set theory and then landuse classification rules are established. At last,according to the error matrix analysis and sampling analysis, the overall accuracy of region multi-center method based on fuzzy rough set theory is 88% and the Kappa coefficient is 0.8473. Compared with maximum likelihood methods and neural networks method,both of these values have been proved.This result indicates that, the fuzzy rough set theory can solve the uncertainty with remote sensing classification,and region multi-center method based on fuzzy rough set theory is an effective classification method,and that it can improve the classification accuracy of RS image.