Researches On Hyperspectral Image Recovery Algorithm Based On Compressed Sensing

Since the second half of the20th century, remote sensing technology has been deeply developed,in the theoretical, technical, or in the application. In which the most prominent is undoubtedly theemergence and development of hyperspectral imaging technique. Hyperspectral images get a spatialinformation, obtaining the spectral information at the same time. It has been widely used in theaerospace, geological exploration, environmental monitoring and other fields. But, with the higherspectral resolution, more imaging, the amount of information becomes more and more. so muchdata have brought tremendous pressure in the calculation, storage and transmission of the data.Therefore, The research of hyperspectral image information processing and compression codingtechnology is much meaningful in the reality.Firstly, a detailed overview about hyperspectral image is introduced, and the hyperspectralimaging principle and its image characteristics is analyzed. Secondly, a research on compressedsensing theory is introduced, with analyzing deeply on the various components of compressedsensing, and point out several potential development direction of compressed sensing in the future.Finally, several hyperspectral images compressed sensing reconstruction algorithm is proposed,combining several characteristics of hyperspectral image. The idea of measurement ratepre-allocation is put forward to aim at the uneven distribution of image information, forming ablocked compressed sensing based on measurement rate pre-allocation. Hyperspectral imagesequences has strong spectral correlation. Researching on the impact of the correlation for thereconstruction of the iterative process, we have made deeply research on two aspects ofhyperspectral image compressed sensing: According to the strong spectral correlation, the idea oflinear filtering between bands to enhance the quality of reconstruction is proposed. Others, in termsof apparent correlations between the image series, a spectral decorrelation model is established. Thespectral redundancies are removed in the measurement data of current image by the model, and thedecorrelation image data is more sparser, which can be reconstructed easily.The main work of this paper is to combine Hyperspectral image compression with compressionsensing theory, several hyperspectral image compressed sensing reconstruction algorithms areproposed. The followings are main content:(1) A image block compressed sensing algorithm based on measurement rate pre-allocation isproposed. Which is cutting up image into block, and calculating the estimate entropy of eachimage block to control the measurement rate pre-allocation. Experiment results show that thereconstruction quality is greatly improved, and the complexity of the algorithm is essentially the same.(2) A hyperspectral image compressed sensing algorithm based on the linear filtering between bandsis proposed. In the algorithm, current results in the each iteration and previous framereconstruction wavelet coefficients make the linear filtering between bands. The new iterationstop condition is designed by reconstruction coefficients of the previous frame. Experimentresults show that the reconfiguration time is greatly reduced, and the reconstruction quality hasimproved significantly.(3) A hyperspectral image compressed sensing algorithm based on decorrelation model is proposed.In terms of apparent correlations between the hyperspectral image series, the spectral correlationredundancies are removed in the measurement data of current image by establishing thedecorrelation model in the reconstruction process. The decorrelation image data has smallerentropy, which can be reconstructed easily. The experiment results can be seen that thereconstruction quality improved significantly.