Research On Compressed Sensing Reconstruction Method Of Hyperspectral Remote Sensing Image Based On Space Spectrum Combination

Due to the limitation of the capacity of data storage and transmission,hyperspectral remote sensing images containing a large amount of spatial spectrum information can not be transmitted quickly and effectively.At the same time,the acquisition of hyperspectral remote sensing images will be interfered,making the collected hyperspectral remote sensing images will produce noise interference information.Therefore,it is necessary to study the hyperspectral remote sensing image itself and its acquisition and reconstruction.Different models or methods are used to optimize and upgrade the hyperspectral remote sensing image at the acquisition and reconstruction ends.In this paper,the hyperspectral remote sensing image itself is elaborated in detail,and the prior information and its own characteristics of hyperspectral remote sensing image are analyzed.At the same time,combined with the characteristics of hyperspectral remote sensing images and their representation in different modes,the compressive sensing reconstruction methods related to hyperspectral remote sensing images are compared and studied.The reconstruction algorithms proposed in the previous literature and the improved reconstruction algorithms do not fully take into account the spatial correlation and inter-spectral correlation of hyperspectral remote sensing images themselves,so they will be reconstructed on the basis of errors,which will lead to the expansion of the impact of errors.In view of the problems arising from the acquisition and reconstruction of hyperspectral remote sensing images,the characteristics of hyperspectral remote sensing images and image reconstruction algorithms are studied.The main research work of this paper is as follows:(1)The data model of hyperspectral remote sensing image is analyzed and studied,which is mainly divided into two kinds of data models:one is multi-band image aggregate model,which decomposes the whole image into multi-band based on the construction of the image itself,and makes full use of spatial-spectral correlation to solve the problem in the reconstruction process;the other is linear mixed model,which is based on the image itself.Starting from the smallest component pixel,the pixel is decomposed into endmember and abundance,and the reconstruction image problem is transformed into solving more sparse abundance matrix problem,which improves the accuracy of reconstruction image.(2)Constructing a linear mixing model of space-spectrum combination,extending the two-dimensional end element extraction into multi-dimensional space,a method of hyperspectral remote sensing image reconstruction based on space-spectrum combination and linear mixing model is proposed.The characteristics of hyperspectral remote sensing image are represented in the linear mixing model,and the end element is extracted by space-spectrum combination method,and then the weight of hyperspectral remote sensing image is carried out.Structure.Experiments show that this method not only improves the reconstruction effect,but also reduces the noise impact on the premise of low complexity.(3)Based on the spatial correlation of hyperspectral remote sensing images,a new method of band classification for hyperspectral remote sensing image reconstruction is proposed.The bands of hyperspectral remote sensing images are classified according to the influence degree of noise,and the bands with strong noise interference are reconstructed by decreasing unilateral prediction.The bands with weak noise interference are reconstructed by bilateral prediction,and the non-interference bands are reconstructed by iteration correction of reconstruction residuals combined with spatial correlation.The spatial spectrum characteristics of hyperspectral remote sensing images are fully utilized.Experiments show that the effect of image reconstruction is greatly improved.