Functional Representation And Classification Of Hyperspectral Images Based On Spatial And Spectral Correlation

With the deepening of hyperspectral remote sensing technology and application,people can obtain hyperspectral images with higher spatial and spectral resolution.How to effectively use the spatial and spectral information in images to improve the classification results has important research significance.Traditional hyperspectral image classification methods treat hyperspectral data as vectors or tensors,and fail to fully mine the continuity and correlation characteristics of spectral vectors,which leads to challenges such as high dimension,spatial and spectral redundancy and nonlinearity in hyperspectral image classification.Therefore,based on functional data analysis and combined with the characteristics of hyperspectral data,this paper proposes a functional representation method of hyperspectral images using spatial and spectral correlation.The main research work is as follows:(1)Aiming at the existing hyperspectral classification method based on functional data analysis can’t effectively by using hyperspectral image rich texture and shape of space such as the characteristics of the problem,based on very pixel contains pixels with similar characteristics of the assumption,this paper proposes a hyperspectral images based on spatial correlation of pixels constraint function representation.Firstly,hyperspectral data are expressed in functional form by linear combination of basis function system.Then,Entropy Rate superpixel Segmentation(ERS)is performed on the first principal component of hyperspectral images to obtain the spatial neighborhood relationship of pixels.Finally,the spatial coherence term is introduced into the sum of squares of fitting error term to realize the superpixel constraint.The fitting results demonstrate the effectiveness of the proposed method.(2)In order to make full use of the high correlation between adjacent bands of hyperspectral images,a functional representation method of hyperspectral images based on spatial and spectral correlation is proposed.Using the above superpixel constrained functional representation model,the hyperspectral data are represented by segmented functions according to the spectral correlation coefficients.On the basis of the above Functional representation,Functional Principal Component Analysis(FPCA)was used to extract the Functional features of the spectral curve fitting function.Combined with Support Vector Machine(SVM)classifier,hyperspectral images are classified,and comparison algorithms are designed to verify the effectiveness and advantages of the proposed method.