Research On Hyperspectral Image Reconstruction Algorithm Based On Compression Sensing

Hyperspectral images with the advantage of combining maps are widely used in medical diagnosis,agricultural detection and other fields.However,the huge data collection makes it difficult to measure,transfer and rebuild.The emergence of compression sensing theory brings new ideas for compression and reconstruction of hyperspectral images.In this paper,compression sensing technology is applied to the sampling and reconstruction of hyperspectral images,and its compression reconstruction algorithm is studied.The specific research contents are as follows:Firstly,in the spatial dimension,in order to solve the problem that the quality of reconstructed image is poor due to the same measuring matrix used by traditional algorithms on the measuring end,a compressive sensing hyperspectral image reconstruction algorithm based on adaptive block is proposed.In this method,twodimensional image Entropy is taken as a measure of texture details of hyperspectral image,and the image is divided into adaptive blocks by texture details distribution of the image.Different measurement values are allocated to different image blocks according to the image Entropy and weight of the sub-blocks for compressed sensing reconstruction.Secondly,on the spectral dimension,an adaptive sampling hyperspectral reconstruction algorithm based on band grouping is proposed to solve the problem that the global reconstruction cannot accurately reconstruct the intervals in the information set due to the uneven distribution of information in the spectral intervals.This method segments the spectral intervals according to the information entropy between the spectrums,and improves the maximum entropy threshold algorithm.The previous segmentation threshold is used as the initial value of this cycle to achieve multi-threshold segmentation for further processing of non-smooth intervals.Finally,in order to solve the estimated value directly by the least square method in each iteration of Compressed Sampling Matching Tracking(CoSaMP),which severely limits the overall operational efficiency of the reconstruction algorithm,a new reconstruction algorithm of CoSaMP compressed sampling based on conjugate gradient improvement is proposed.In the improved algorithm,the conjugate gradient method is used to solve the least squares problem in CoSaMP to reduce the computational pressure brought by the least squares solution when generating largescale matrix from hyperspectral images.The experimental results show that the PSNR of the spectral image reconstructed by the adaptive blocking algorithm is improved by 2-4 dB,the SSIM is improved by0.27,and the visual effect of the reconstructed image is also improved significantly.The spectral fidelity of the band grouping method is improved significantly,the single threshold piecewise reconstruction PSNR is improved by about 2 dB,and the spectral reflectance ratio norm and the reconstruction time are also reduced.The PSNR result of double threshold reconstruction is improved by about 3.6 dB and the spectral reflectance ratio norm is reduced to about one third.With the improvement of the CoSaMP algorithm,the running time of the algorithm is reduced by 10-15 seconds,and the improved algorithm will reduce the running time without reducing the reconstruction accuracy of the original algorithm for hyperspectral images.