Hyperspectral Remote Sensing Image Classification Based On ELM And RBFNN

With the development of modern remote sensing technology,hyperspectral remote sensing has become a hot topic in the field of remote sensing technology.It is the abbreviated form of the high spectral resolution remote sensing.Hyperspectral data have a lot of spectral bands,and meanwhile the spectral resolution is very high.The target image obtained by imaging spectrum technology contains three kinds of information,such as abundant space,radiation,spectra.It provides more accurate spectral information than multispectral sensor,so the classification of remote sensing image for object recognition is greatly improved.This implies that it has a broad application prospect in civilian and military.However,the hyperspectral remote sensing image is with characteristics of bands,large amount of data and the uncertainty of data.It is also easily affected by Hughes phenomenon.Therefore,it is also a challenging problem how to apply it to solve many applications.With the rapid development of computer science,more and more spatial data are stored in spatial database.It is difficult to process these huge data by traditional geographical information system technology.As a result,the spatial data mining technology is introduced to study the geographic information system.One of application of spatial data mining to geographic information system is to deal with remote sensing image data in the form of artificial intelligence for remote sensing.Currently,many spatial data mining algorithms are successfully applied to the remote sensing image processing.Along this direction,two problems are investigated in this thesis:According to the high dimension characteristics of hyperspectral remote sensing image data,it is necessary to extract the valuable features as far as possible by dimensional reduction.One here first uses the linear discriminant analysis method to reduce the dimension of hyperspectral remote sensing image data.Then hyperspectral remote sensing image classification can be done by adopting the method of extreme learning machine.To assess the effectiveness of the proposed method,the Aviris hyperspectral remote sensing image is used to do the experiment.The accuracy of classification is high.The experimental results show that the algorithm has higher classification accuracy.To retain as much as possible of valuable information of Hyperspectral remote sensing image,the linear discriminant analysis method is carried out to deal with the problem of dimension reduction of hyperspectral remote sensing image data.The radial basis function(RBF)neural network method is applied to classify these data.To prove the effectiveness of the proposed algorithm,the experiment uses the Salinas-A hyperspectral remote sensing image data.The accuracy of experiment result is high.The training convergence speed is fast.It has good effect of classification.The experimental results prove the feasibility of this method.