| When remote sensing technology is more and more strict with spatial spectral resolution of hyperspectral images（HSI）,the burden for spectral imaging equipment to sampling,transfer,store is heavier and heavier.It is indisputable that the further development of spectral imaging technology has been hampered.Compressed sensing theory,which realizing the synchronization of signal acquisition process and compression process and accurately recovering the original signal from a small number of measurement sampling data points,provides a new research direction for the compression and reconstruction of HSI.HSI has three-dimensional information features.According to the data characteristics of HSI,it is a key for increasing the reconstruction accuracy to design a new reconstruction strategy or algorithm.A reconstruction algorithm based on dual-band prediction and compressed sensing is introduced.This article’s principal research is as follows:（1）For the strong inter-spectral correlation of hyperspectral images,a reconstruction framework for HSI based on dual-band prediction is mainly introduced.At the sampling,according to the high correlation between the spectra of HSI,the K-means clustering algorithm is introduced to group the hyperspectral images.And the cluster center image and the band image with the largest sum of SSIM values of other bands in the group are used as the double reference bands of each group;At the reconstructing,the SL0 algorithm is used to reconstruct the dual reference band images of each group independently;then the reconstruction value of the dual reference bands and non-reference band are observed the difference and reconstruct the band respectively,to obtain the initial predicted value of the non-reference band;Then,the initial the predicted value is independently iteratively corrected until the mean square error of the two iterations is less than the set threshold;Finally,the nonreference band is obtained by weighted average fusion.The experimental data is obvious: the reconstruction accuracy of this algorithm is significantly better than only independently reconstruction or other existing inter-reconstruction algorithms.（2）For the reconstruction algorithm of compressed sensing,a reconstruction algorithm based on the CSH-SL0 algorithm is mainly introduced.Based on the dual-band prediction model between spectra,the algorithm mainly improves and optimizes the SL0 algorithm in three aspects: continuous smooth function,iterative method,and step size update method.Based on the analysis of the "steepness" degree of the existing continuous smooth function a more "steep" compound sinehyperbolic function is constructed to approximate the L0 norm of the sparse signal;And BFGS quasi-Newton method is introduced for renewing the descending direction.It is a way to renew iteration step by the one-dimensional Armijo-Goldstein search.Without changing the previous grouping strategy and prediction model,CSH-SL0 algorithm is used as a new hyperspectral compressed sensing reconstruction algorithm.The experimental data is obvious: this algorithm is significantly improving in the details,visual effects and so on. |
