The Study Of The Partition Of High Spectrum Mixed Pixel And The Topographical Objects Classification

With the deep development of quantification, there were wide concerns on the hyperspectraltechnology of remote sensing both in the home and abroad because of the high spectral resolution.It obtained many achievements on the applications of geology, vegetation ecology, soil, and cityplanning and has become an important developmental direction on current remote sensing.Regarding the achievements, the applied issues on the classification methods of topographicalobjects hyperspectral remote, the selection of training samples and the decomposition of mixed-pixels have also get abroad concems which have gained quite great progress.The thesis developed with the EO-1 hyperspectral remote sensing data, carried on theselection of wave band, transformed the absolute radiation value, repaired the defective wire,removed the stripes, corrected the atmosphere and pretreated the geometry. It applied several kindof unsupervised and supervised classification methods, gained the type 2 investigations data in thehigh spectra pictures area, and matched with the high spectrum picture, inspected the classificationprecision by this,. As a result, the highest classification precision achieved 79. 8954ï¼…. On the basisof the work above, it attempted to classify the topographical objects hyperspectral remote sensingdata with the combination of the traditional classification method (the ISODATA method in theunsupervised classification and the minimum distance method and the maximum likelihood methodin the supervised classification) and the method based on spectrum characteristic classification (thespectral angle mapper), the precision achieved 81.2231ï¼….This thesis carried on the MNF transform to the picture after high spectrum pretreatment(theminimum noise fraction), obtained the bigger picture including an information content, andclassified the hyperspectral remote sensing data using the ISODATA method in the unsupervisedclassification, the minimum distance method and the maximum likelihood method in the supervisedclassification and the spectrum angle mapping method. It produced the pixel purity index pictureusing the inside dimension of the data, carried on the N dimensional visualizer and carried on thelinear unrestraint decomposition to the pixel using the linear model on this basis of MNF transform.It concluded that the classified precision of the spectrum angle mapping in the supervisedclassification is higher through comparing the unsupervised classification with the supervisedclassification. It concluded that the ISODATA classified precision after the spectrum angle mappingwas higher through comparing the ISODATA classification, minimum distance classification andthe maximum likelihood classification after the spectrum angle mapping with the spectrum anglemapping. It also concluded that the classified precision of the linear model was highest throughcomparing the unsupervised classification with the supervised classification and the linear model.