Multiclass Sorting Research On Hyperspecral Based On Supporting Vector Machine(SVM)

Earth remote sensing is data collection on the environment, geology, climate, andother characteristics of the Earth by means of sensors positioned in the air or in Earthorbit. The data of hyper-spectrum has many bands and has high spectral resolution.Generally there are about more than two hundred bands from visible light to infrared.If very highly rise the spectral resolution, it will be possible to sense earth-objectsfinely and distinguish the similar objects. So hyperspectrum is widely used.The hyperspectral image classification is to assign a class mark to every pels ofhyperspectral images. The general hyperspectral sorting arithmetics relate to twofactors: character extractiving and choosing, and sorter designing. We focus ourresearch on the strategy of multi-class classification based on the support vectormachine (SVM), proposed an improved multi-class classification method..Based on the analysis of current multiclass strategies, an―eliminationgame‖–liked decision tree method Elimination Decision Tree (EDT) is proposed andgeneralized to solve the multiclass SVM problem. Experimental results show that itcan keep the high accuracy of―round robin‖one-against-one (OAO) method, but onlyneeds about50%test time of OAO. To obtain a much faster multiclass strategy,according to the two factors that directly affect the classification, i.e., the number oftwo-class problems and the number of support vectors, a Fast Binary Tree (FBT) isproposed. Experimental results show that it only needs about25%test time of OAO,but exhibits less than0.5%descended classification accuracy.