The Application Of Independent Component Analysis In Land Cover Information Extraction With Remote Sensing Image

In recent years, the land cover information of the cities and their surroundingareas keeps changing frequently in the rapid urbanizing process. So from theperspective of the urban land use and the prevention of disasters, how to effectivelymonitor these changes has always been the key researching direction for scientificand technical workers. Besides, with the development of remote sensor technology,remote sensing (RS) has been able to acquire a good deal of land cover information;however, how to extract the precise land cover information out of these RS databecomes a hot and difficult issue in the monitoring of land cover changes.As a kind of blind source separation (BSS) technology, independent componentanalysis (ICA) is able to separate the source signals from the observing mixedsignals. Support vector machine (SVM) can nonlinear map the sample data which ishard to be linear separated in low dimensional space to high dimensional featurespace, and finally fulfill the nonlinear classification and extraction of the sample dataaccording to the structural risk minimization principle (SRM). They both havetremendous applicable potentials in RS image land cover information extraction.This paper has delved into ICA, SVM and their application in RS image land coverinformation extraction, which are mainly embodied in the following aspects:(1) Aiming at the actuality of ICA model in the extraction of RS image landcover information, this study puts forward the method of variational Bayesian ICA,whose working principle is that through the introduction of Bayesian network intoICA model, work out the posterior probability distribution of different types ofground objects with Bayesian inferences and get the simplification with the aid ofvariational approximate algorithm, so as to make the independent componentsseparated from ground objects approach the earth surface real distribution as muchas possible. RS image analyzing results indicate that variational Bayesian ICA hasthe advantages of good stability, high separation degree, and good visual effect. (2) On the basis of a comprehensive consideration of variational Bayesian ICA,SVM algorithm and RS image features, this study puts forward a RS image landcover information extraction method with the combination of variational BayesianICA and SVM. The extraction result shows that the combined method is equippedwith a high noise immunity, universality, extraction precision and visual effect.(3) This paper tries to apply the combination of variational Bayesian ICA andSVM into the study of the RS image land cover information extraction of Chongqingurban core areas, and has extracted each type of the land cover information in1988and2007respectively, and has given an analysis of their temporal&spatial variation.It turns out that the urban construction land is mainly expanding toward the northeast.The extraction principle and technique process in the application of RS image landcover information extraction has great referential significance for the land coverinformation extraction and the study of the changes in other hilly cities.