Research On Compression And Reconstruction Method Of Hyperspectral Image Based On Grouplet Transform

Hyperspectral imagery has a large amount of redundant information due to its spatial spectral characteristics,which is difficult to use and store.When it is used as a research object in scientific research and application fields,the speed of data processing is also reduced due to the large amount of data.Longer and less efficient is a problem in processing hyperspectral images.After the image is compressed,the data is stored in a small amount of information,and the original signal can be accurately approximated from the information,thereby solving the problem that the hyperspectral image storage amount is large,and it is difficult to calculate and operate.For this problem,This paper starts research with the support of the National Natural Science Foundation of China(No.51675258,51261024),the Science and Technology Project of Jiangxi Provincial Education Department(No.GJJ150699).This paper proposes a series of compression reconstruction algorithms based on Grouplet transform,and sparse compression and accurate reconstruction are the main purposes of this paper.In this paper,the superiority of different algorithms is compared through experimental research.Then,these algorithms are applied to natural hyperspectral images to verify the correctness and reliability of the experimental results.Under the compression and reconstruction of different algorithms,some valuable conclusions are obtained:First,the hyperspectral image is sparsely expressed by the Grouplet base and the Wavelet base,and then the inter-spectral redundancy is compressed by the AKL transform.Grouplet transform is superior to Wavelet transform in terms of sparse capability of hyperspectral images.Some of the bands affected by noise in the hyperspectral image can be better represented by the sparse process of Grouplet transformation.The sparse coefficients expressed by the high-quality sparse base can provide a high-precision sparse model for the compression reconstruction algorithm.Therefore,the Grouplet-AKL algorithm significantly improves the inter-spectral compression and sparse reconstruction accuracy of hyperspectral images,and the accuracy of the algorithm is better than the Wavelet-AKL transform.Second,after the spectral inter-spectral redundancy is compressed,the resulting data needs to continue to remove the redundancy of the spatial domain.The Grouplet transform is combined with the compressed sensing algorithm to form a Grouplet-CStransform.In this paper,the algorithm is applied to the sparse compression of hyperspectral images,and the two most classic greedy reconstruction algorithms are used for information recovery.By compressing and reconstructing hyperspectral images and then comparing the performance of different sparse bases and reconstruction algorithms,the following results are obtained:(1)Compared with the Wavelet basis,the Grouplet-based error is significantly reduced for each single-band sparse reconstruction of hyperspectral big data,which indicates that the Grouplet transform can provide better sparse expression and image reconstruction accuracy;2)The reconstruction effects of the omp and romp reconstruction algorithms are ideal,the reconstructed image has a clear outline,the texture details are clear,and the recognition is high.At the same sampling rate,the average signal-to-noise ratio(APSNR)of the romp reconstruction algorithm is significantly higher than the omp reconstruction algorithm,and the recovery accuracy of the two reconstruction algorithms decreases significantly with the increase of the spatial compression ratio.For the overall sampling compression of the hyperspectral image in the 3D spatial domain,the Grouplet basis is used for 3D sparse expression,which not only improves the compression rate,but also improves the reconstruction quality of the hyperspectral image.Third,the hyperspectral image in the actual problem contains a complex type of noise,and the compressed sensing algorithm fails to estimate the noise signal,so that the compressed data is affected by the noise,thereby affecting the reconstruction accuracy.In this paper,the Bayesian idea is introduced into the compressed sensing algorithm.The hidden noise signal is estimated while the sparse model is projected,and the Bayesian compressed sensing algorithm is combined with the Grouplet transform.The Grouplet-BCS algorithm is proposed to improve the signal reconstruction accuracy.In the signal information reconstruction process,the sparse signal data is accurately restored according to the posterior distribution of the noise.Considering the shortcomings of Bayesian compressed sensing algorithm,a variational Bayesian compressed sensing algorithm and a fast Bayesian Compressed Sensing algorithm are proposed and combined with Grouplet transform.The Wavelet base and the Grouplet base are combined with six compression reconstruction algorithms respectively,and the sparse performance of different algorithms and the accuracy of compression reconstruction are compared.The results show that:(1)Compared with the Wavelet transform,the natural image is sparsely expressed using the Grouplet basis,and theperformance of the obtained coefficients is gradually optimized with the increase of the number of sampling points,and is closer to the original signal;(2)The Variational Bayesian Compressed Sensing algorithm can provide a better sparse model for the image reconstruction process than the fast Bayesian compressed sensing algorithm,but its computational complexity is large,which makes it take longer to run.The fast algorithm can greatly improve the running speed,and with the increase of the number of sampling points,the accuracy is almost the same as the variational algorithm.Fourth,the natural hyperspectral image is subjected to sparse reconstruction by an adaptive threshold Grouplet transform.The image information and the noise signal of each band are sparsely expressed,and the denoising process is performed by adaptive thresholding,and part of the redundant information is removed in advance,thereby obtaining a better quality sparse model.For the compression reconstruction of hyperspectral image,this paper proposes a compressed sensing measurement matrix of adaptive projection based on Bayesian theory,which not only adaptively estimates the sparse model but also estimates the noise signal.Therefore,the two algorithms are combined to form an adaptive Grouplet-FBCS algorithm,which is used for the compression and reconstruction of natural hyperspectral images.Even if the images of each band are affected by different degrees of noise,the algorithm can adaptively approximate the original signal,accurately reconstruct image information.By comparing the reconstruction error and signal-to-noise ratio peaks of the four algorithms,the results show that:Compared with the other three compression reconstruction algorithms,the algorithm has stronger ability to remove redundant information and denoise from natural images,and the accuracy of information recovery is better.The structural similarity and feature similarity of the reconstructed image and the original image are higher,so that the reconstruction effect is more reliable.