Application Research Of Kernel Sparse Algorithm In The Object Detection For Hyperspectral Images

Hyperspectral remote sensing is a means of obtaining information through fine observation of ground objects.It is the most significant achievement in the history of remote sensing development.Target detection of hyperspectral images is a research hotspot at this stage.In the civil field,hyperspectral target detection has a wide range of applications in crop monitoring,environmental disaster reduction,archaeology,resource exploration,etc.In the military field,hyperspectral target detection has many very successful applications in target anti-camouflage,target reconnaissance,mine detection and so on.Based on the characteristics of hyperspectral data and the dimensionality reduction by band selection,the target detection of hyperspectral image is realized by combining the kernel method and sparse representation theory.The paper first introduces the basic theory related to hyperspectral target detection.The basic characteristics of hyperspectral data are summarized.The important role of data reduction in hyperspectral image is expounded.Two kinds of dimensionality reduction methods based on band selection and feature extraction are studied.The basic theory of hyperspectral image target detection is introduced.The classical algorithm for hyperspectral target detection;at the same time,the related theory of sparse representation is introduced,the mathematical model of signal sparse representation is given,and two important problems of sparse representation are: dictionary training and solving sparse representation coefficients,giving a classic Solution.Then,for the problem of high redundancy of the hyperspectral data and large amount of data,the hyperspectral data is subjected to dimensionality reduction processing.The artificial fish swarm algorithm and the ant colony algorithm are analyzed and improved,and they are combined to obtain a hybrid swarm intelligence band selection algorithm.The main subset of the band is selected from all the original bands,and the band subset has a strong class.It is separable.The experimental results show that the proposed method can reduce the redundancy and correlation of hyperspectral image data while maintaining the complete characteristics of the spectrum.Finally,the overall implementation of the target detection method for hyperspectral images is implemented.Firstly,based on the basic sparse representation,a hyperspectral target detection algorithm based on sparse representation is proposed.Then,based on the nonlinear characteristics of hyperspectral data,the kernel method is introduced into the sparse representation,which has the ability to process nonlinear data,obtain the kernel sparse representation model,and nucleate the dictionary training algorithm and the sparse representation coefficient solving algorithm.The kernel singular value decomposition method(K-KSVD)and the kernel orthogonal matching pursuit algorithm(KOMP).Based on this,a hyperspectral target detection algorithm based on kernel sparse representation is proposed.This method combines the advantages of band selection,sparse representation and kernel method.The experimental results show that the target detection method based on kernel sparse can effectively improve detection.Rate and superior performance.