Research On Hyperspectral Remote Sensing Imagery Classification Algorithms Based On Decision Tree

The rapid development of hyperspectral imaging spectrometer writes a new chapteron the field of hyperspectral remote sensing technology. It has been successfully used ongeological disasters analysis, military target recognition and agricultural production, etc.While the hyperspectral data can not be effective used for its speed of development ondata processing technology. On various hyperspectral processing methods, thehyperspectral classification acts as the basis of other advanced application. Therefore, ithas a great significance of effective used on mining ground information.First, hyperspectral remote sensing technology and classification theory is introducedby this paper. Then it summarizes the characteristics on hyperspectral data. Thecomparative study on frequently-used classification algorithms is the major method in thispaper, and the disadvantages of traditional classification methods is pointed out. On thebasis of these work, the advantages on decision tree based classification algorithm issummarized. Hence, this paper takes the decision tree based classification algorithm asbasic algorithm for further study.Second, through systematic analysis on hyperspectral data, a first-order parametricmodel on homogeneous area and a nonparametric statistical model on nonhomogeneousareas are built by this paper. This successful combination between binary decision tree andmultidimensional scaling develops into a multidimensional scaling hyperspectralclassification algorithm on the basis of multidimensional scaling.In the last part, an experiment is designed. At first, the experiment combinesmultidimensional scaling with spectral information divergence, battacharyya distance,diffusion distance and recursive hierarchical segmentation respectively. And the resultproves that the association measure designed by the author is optimal classificationmethod. Then it makes contrast with support vector machine method and maximumlikelihood method on their classification effect. And it shows that the measure designed bythis paper has the highest classification accuracy.