The Research On The Technology Of Oceanic Land Cover Classification And Accuracy Assessment

In recent years, the frequent appearance of the extreme weather in the global scopebrought enormous influence to countries all over the world. Global change and influencegradually become the focus of attention of the mankind. In order to effectively deal with theglobal change and influence, some countries are dedicated to the study of solutions to globalchanges brought about by the series of scientific problems, global change research isbeginning to be an upsurge. In this hot, China launched the863national projects "Higherresolution Global Land Cover Mapping Project", the research on Oceania land cover andaccuracy assessment technology is based on the project. The difficulty of the Oceania landcover classification and accuracy assessment technology is that the research is bigger, whichleads to large amount of data, complex land cover type. Traditional classification methodssuch as support vector machine, maximum likelihood and non-supervised classification or forlarge feature space or to give less than ideal accuracy can’t meet the requirements of the study.The using of decision tree solved this difficulty. Decision tree classification method is flexible,intuitive, clear, strong, operational efficiency higher characteristic, and has been successfullyapplied to many classification studies, these studies also contains some researches onclassification algorithm of decision tree. C5.0decision tree algorithm is now a classificationalgorithm at the leading edge of algorithm, for some improvement based on C4.5, C5.0algorithm has more challenge.On the basis of studying C5.0algorithm, this paper studied the using process of C5.0inthe Oceania land cover classification, and the New South Wales and Victoria for Pacific areawere as the study area, we introduced the using process. At the same time, aimed to theproblem that ETM+images have many mixed pixel, we introduced the spectral unmixingtechnology into the accuracy assessment, by comparing with traditional accuracy assessmentresult, we concluded the result of accuracy assessment based on sub-pixel is more Objectiveand credible.