Research On Dimensional Reduction And Classificasstion Method Of Hyperspectral Remote Sensing Images

Hyperspectral remote sensing is a new earth observation technology which is started in1980s.The hyperspectral image has a characteristics of high spectral resolution.It provides a more wealth of details information for Ground objects,are Widely used in agriculture、 forestry、 geology、 the sea、 the military、 and other fields.However,The large quantity of hyperspectral data has caused a few inconvenience for a variety of applications, so it is very important to develop fast and accurate methods to extract useful information from the large data in hyperspectral image is the premise of the application.This paper started with analyzing the existing hyperspectral image processing technology.Take Xiaotangshan ChangpPing BeiJing PHI data as example. Proposing a new method which combine Ant Colony Optimization algorithm(ACO)、 Independent ComponentAnalysis(ICA、 Supptort Vector Machine(SVM) for hyperSp-ectral image processing.The Main work as follows:(1) Systematically summarized the existing data dimension reduction method and image classification method. Dimension reduction of hyperspectral images includes band selection and feature extraction.The band selection includes method based on information standards and the separability between two difference classes, feature extraction comprise ICA、 Principle Component Analysis(PCA)、 Minimum Noise Fraction(MNF),and so on. the existing classification methods are consist of Minimum Distance classification、Maximum Likelihood classification、 Neural network classification、 SVM,ect.(2) This paper focuses on the concepts of ACO and ICA.Proposing the ACO-ICA method.Which utilized ACO method divided all bands into four different subspaces,and proceed ICA transformation in each subspaces.Then we got four subspaces,and through the ICA we extract the feature component which contains higher eigenvalue.Combinating the new feature from each subspaces to obtain new data with the lower correlation and the huger effective information.(3) Particularly introduction the principle of SVM. Proposing the SVM Classification based on ACO-ICA.Compared the consequence of SVM Classification whice based on ACO-ICA with the results of SVM Classification based on the whole spaces ICA.At the same time,Compare the consequence of SVM Classification with traditional Classification method.The results show that:The result of this paper got a overall Accuracy of85.28%,higher than other Classification method.This proved the feasibility of the method this paper proposed.