Research On Hyperspectral Image Classification Method Based On Sparse Representation

Hyperspectral remote sensing is a kind of high spatial resolution and high spectral resolution of remote sensing technology,the wavelength range from between visible light and infrared light,can from hundreds to thousands of narrow and continuous spectrum band image capture mining,is conducive to accurately interpretation to similar features and is widely used in agriculture,military,geological exploration,Marine monitoring and other fields.The rich spectral information and two-dimensional spatial information of hyperspectral remote sensing image are integrated into one,which can easily distinguish ground objects.However,its high feature dimension and small sample characteristics also bring great challenges to accurate hyperspectral classification.In order to improve the classification accuracy of hyperspectral remote sensing images and summarize the research achievements and shortcomings of outstanding scholars represented by Y.Chen,this paper proposes three hyperspectral remote sensing image classification methods based on sparse representation:1.A hyperspectral image classification method combining 3D fast rolling guided filtering and sparse representation.Firstly,the hyperspectral image is dimensionalized,and the first principal component is used as the guide map,and the effective spatial feature extraction is achieved by combining the three-dimensional fast cyclic guided filtering algorithm.Then,the spectral feature and spatial feature are linearly fused,and finally,the sparse representation method is used to classify the hyperspectral image.It is verified on three standard data sets that the proposed method has higher classification accuracy and better edge preservation effect;2.A hyperspectral image classification method based on improved discriminant sparse neighborhood preserving embedding.The samples of each class of hyperspectral image are regarded as a manifold,and the optimal projection is sought to maximize the inter-manifold scattering and minimize the intra-manifold scattering of the data and maintain the sparse structure of the entire data.The coding distinguishing information in the multiple structure has a good classification and discrimination ability.The proposed method is validated on three standard data sets to improve the classification accuracy of hyperspectral images3.Hyperspectral image classification method based on sparse representation of spacespectral kernel.First,the huge hyperspectral image is divided into several blocks according to the column,then the corresponding large matrix is built and the spatial similarity is measured by neighborhood filtering in the kernel feature space,and finally the index is carried out by the appropriate kernel items,which solves the convex optimization problem of each tested pixel.It is verified on three standard data sets that this method is better than other kernel sparse representation classification methods.