Hyperspectral Target Detection Based On Sparse Representation

Hyperspectral target detection has many applications in the civil and militaryarea. Typical examples include the detection of specific terrain features andvegetation, mineral, or soil types for resource management; detecting andcharacterizing materials, surfaces, or paints; the detection of man-made materials innatural backgrounds for the purpose of search and rescue; the detection of specificplant species for the purposes of counter narcotics; and the detection of militaryvehicles for the purpose of defense and intelligence. The purpose of this article is topropose a new algorithm for target detection in hyperspectral imagery (HSI), whichtaking into account of the specialty of hyperspectral imagery. And it can promote thedevelopment of hyperspectral remote sensing.The objective of this article is to study the hyperspectral target detectionalgorithm based on the sparse representation including the following three aspects:First of all, the questions based on the sparsity model about how to design adictionary and how to obtain the coefficient have been studied. Then considering thecharacteristics of hyperspectral imagery, the sparsity model for hyperspectralimagery has been studied. Considering the spatial and spectral characteristics ofhyperspectral imagery, the spatial-spectral constraint model including the jointsparse constraint model and the Laplacian constraint model and the3D-DWT-ICAconstraint model.And then the conventionally matched subspace detector (MSD) algorithm hasbeen studied. After that the new target detection algorithm based on the sparserepresentation has been studied. Substituting SR for the conventional subspacemethod, a sparse matched subspace detector (SMSD) is developed. Moreover, thespatial-spectral constraint model is exploited to extract the spatial and spectraldistribution in the hyperspectral imagery and capture the joint spatial-spectralsparsity structure. Experiments are conducted on real hyperspectral data. Theexperimental results show that the proposed algorithm outperforms thestate-of-the-arts sparse detection algorithm.Finally, in order to solve the nonlinear characteristic of hyperspectral imagery,the kernel methods have been considered. Firstly, we have studied the questionabout how to obtain the coefficient of kernel sparse representation. Then the targetdetection algorithm for hyperspectral imagery based on kernel sparse representationhas been studied. And at last we present a kernel realization of a sparse matchedsubspace detector (SMSD) that is based on sparse representation model defined in a high-dimensional feature space associated with a kernel function.