Research On Reconstruction Of Hyperspectral Remote Sensing Image Based On Compressed Sensing

In recent years,along with the rapid development of remote sensing technology,hyperspectral technology has made great progress.Hyperspectral remote sensing images can provide rich spectral information and spatial information of the observed objects.At the same time,the large amount of spectral data brought by high spectral resolution has also caused a series of problems in the transmission and storage of hyperspectral imaging technology.The appearance of compressed sensing combines the sampling and compression processes to sample the measurement data far below the sampling rate required by the traditional Nyquist sampling theorem,which greatly reduces the impact of hyperspectral remote sensing images.Sample data volume.In view of the above problems,this paper combines the strong correlation between hyperspectral remote sensing images and studies the reconstruction method of hyperspectral remote sensing images based on compressed sensing.The main research contents and work of the paper are as follows:(1)A reconstruction method of hyperspectral remote sensing images based on group sampling is proposed,which effectively improves the reconstruction effect of hyperspectral remote sensing images.According to the strong correlation between spectral features of hyperspectral remote sensing images,the images are divided into several band groups.Each group defines a reference band,and the corresponding reference band information is added to the reconstruction of non-reference band.Experimental results show that the reconstruction method of group sampling improves the reconstruction effect of hyperspectral remote sensing images to some extent.(2)In the packet sampling reconstruction method,the strong correlation between hyperspectral image spectra is not fully utilized,resulting in poor reconstruction of non-reference band.To solve this problem,a dual-reference band prediction model based on the correlation between hyperspectral spectra was proposed,which further improved the reconstruction effect of non-reference band.Find the two bands in each band group that have the greatest correlation with other bands as the reference band of the group.Based on the model established by the two reference bands,predict the non-reference bands.Experimental results show that the dual reference band reconstruction method improves the reconstruction effect of non-reference band.(3)In the dual reference band prediction model,the sampling rate of the reference band is low,resulting in poor reconstruction.In order to solve this problem,a compressed hyperspectral remote sensing image reconstruction method based on compressive sensing combined with an inter-spectral multi-directional prediction model is proposed,which improves the reconstruction effect of the reference band.First,a reference band is determined for each group,and an inter-spectral multi-directional prediction model is established based on the reference band within the adjacent group to calculate the predicted measurement value of the non-reference band;then,the difference between the actual measured value and the predicted measured value is reconstructed.Finally,it uses the resulting difference vector to iteratively update the predicted measurements until the original image of the band is restored.The experimental results show that the compressive sensing-based hyperspectral image reconstruction method proposed in this paper combined with multi-spectral inter-spectral prediction model can effectively improve the reconstruction effect of hyperspectral remote sensing images under the condition that the sampling rate of images is greatly reduced.This will provide new ideas for the reconstruction of hyperspectral images in the future.