Research On Space-spectral Joint Reconstruction Algorithm For Hyperspectral Images

With the development of information technology,data forms are gradually diversified.In addition to traditional signals,texts,images,videos and so on,some high-dimensional data forms are well known by more and more people.As a typical form of remote sensing images,hyperspectral images have been widely used in water quality inversion,surface feature classification,biopharmaceutical and other aspects.They often play an irreplaceable role in quantitative remote sensing image analysis and feature classification.They have rich spectral reflection bands and can better retain the physical characteristics of the photographed target.However,a hyperspectral image stores the same ground objects’ information of plenty spectral bands,which leads to great pressure on the hardware equipment in the process of hyperspectral image data acquisition,transmission and storage.Therefore,when processing hyperspectral images,how to use the mutual information of multiple bands and remove redundant bands and noise bands is the main research direction of hyperspectral bands.Compressed sensing technology can collect a few of sparse data during sampling and recover it during reconstruction.The sparse data can be recovered through reconstruction algorithm,which fundamentally alleviates the bandwidth and storage pressure caused by the huge amount of data.Aiming at a series of commonness and characteristics of hyperspectral image in compressed sensing,this paper studies the correlative methods,in order to reduce the redundancy of this mission and improve the accuracy of image reconstruction.The main work of this paper is as follows:In order to improve the reconstruction accuracy of hyperspectral compressed sensing,a compressed sensing reconstruction algorithm based on space spectrum combination is proposed in this paper.In this paper,the effects of reconstruction error and sparsity on the image compression sensing accuracy are comprehensively considered,and the spatial characteristics and inter spectral characteristics of hyperspectral images are integrated,so as to build a highdimensional multi-objective atomic index set optimization model.There are abundant redundant bands(spectra)in hyperspectral images,which have great influence on the reconstruction accuracy in compressed sensing spacespectrum joint reconstruction.To solve this problem,a de redundant spacespectrum joint reconstruction algorithm for compressed sensing of hyperspectral images is proposed in this paper.Firstly,based on the joint space-spectrum reconstruction,considering many factors affecting the similarity of hyperspectral images,a multi-target band selection model is constructed to realize the adaptive band selection of hyperspectral image compression sensing;Then,a joint spacespectrum reconstruction algorithm of hyperspectral image based on block compressed sensing is designed.Because there is no unified standard for the measurement of hyperspectral image similarity,most algorithms often measure the similarity between hyperspectral image bands by calculating the mutual information or covariance matrix between bands,but do not make efficient use of the characteristics of hyperspectral image.Based on the above problems,a new similarity measurement method is proposed in the joint space-spectrum reconstruction of hyperspectral images,and a double-layer band selection model and corresponding algorithm are designed.Firstly,the hyperspectral image is decomposed by tensor to realize unmixing,and the spatial and spectral features of hyperspectral image are separated to measure the similarity between bands;The first layer is used as the optimization problem,and then the second layer is used as the selection tensor.