Hyperspectral Remote Sensing Image Classification Based On Relevant Vector Machine

Hyperspectral remote sensing imagery which has strong practicability by virtue of high spectral resolution, huge spectral information and high relevance of adjacent band,is praised by experts and scholars and becoming popular in the field of remote sensing.In order to make up disadvantages in SVM,M.E.Tipping put forward a new kind of supervised machine learning algorithms who based on bayesian statistical learning framework in 2001-- Relevance vector machine(RVM).We can obtain a sparse solution that based on kernel function by probability distribution of forecast through regression estimate in this method. Using the sparse solution we can deal with regression and classification problems. But we find that the classification accuracy of RVM is not high when apply to hyperspectral data,so a novel classification method based on RVM is presented in this paper.Compared with the traditional support vector machine(SVM) classification algorithm in the high dimensional and large sample data we found that the relevance vector machine in Kappa coefficient and overall classification accuracy is worse.Therefore, this paper focuses on the research on the preprocessing stage. Firstly,principal component analysis is used to dimension reduction,the experimental results show that the method can not improves classification accuracy than RVM. Secondly,the linear discriminant analysis method is used to dimension reduction,the experimental results show that the method still can not improves classification accuracy than RVM.Finally,this paper presents an improved relevance vector machine algorithm. The method combine principal component analysis and linear discriminant analysis to reduce the dimensionality of hyperspectral data,then the relevance vector machine model is applied to remote sensing image classification. The experimental results show that the proposed method not only improves classification accuracy than RVM,but also extend the ratio between inter-class distances and intra-class distances.Aiming at the problem that the classification accuracy of high dimensional data is not high, this paper presents an improved relevance vector machine algorithm. Themethod not only keep the main information of the image, but also reduce the dimension and increase the radio between the inter-class distances and intra-class distances,and thus improve the classification accuracy. At last, we analysis the advantage and disadvantage of this method and provide guidance for further research in the future.