Research On Hyperspectral Image Reconstruction Algorithm Based On Compressed Sensing

Hyperspectral image can provide two-dimensional spatial information and provide one-dimensional spectral information of observations,which facilitates the screening of physical properties and structure of form,be widely used in many fields such as space exploration,geological survey and medical research.However,hyperspectral image is different from ordinary two-dimensional image.It is three-dimensional information data with spatial dimension and spectral dimension.The amount of data is huge,which brings great difficulties to various subsequent processing of image data,especially decoding refactoring.The above problems make the development of hyperspectral image difficult,researching the reconstruction of hyperspectral image is of great significance to the development of various fields.As a new signal processing framework,Compressive Sensing(CS)theory has greatly slowed down the pressure of the signal at the encoding.Applying the theory obtain the measured data in directly,greatly reducing the computing power of the data and reducing the pressure on the storage and transmission of data,the sampling of CS is not determined by the frequency band of the signal data,but only with the content and structure of the signal.The birth of CS theory has brought new changes to the processing of signals,igniting the research interests of many reseachers.In this paper,CS theory is applied to the recovery reconstruction of hyperspectral image,and its reconstruction algorithm is studied.The main research of the thesis are:1,Combining the theory of compressed sensing theory and the characteristics of hyperspectral image,a compressed-spectral hyperspectral image reconstruction algorithm based on correlation of space and spectrum is proposed.In the image signal acquisition collection,the hyperspectral image is first grouped by the inter-spectral correlation of each band image;when reconstructing the image,the reference image is first restored by the traditional CS reconstruction algorithm,and the reconstructed reference image is predicted.Non-reference image,and the non-reference image is predicted by the reconstructed reference image,and the calculation is performed.The measured value difference of the non-reference image is predicted,that is,the residual,and then the residual is updated and reconstructed bythe reconstruction method to correct the predicted value.According to the simulation results,compared with the GPSR and SP the similar algorithms,the SL0 algorithm improves the image Indian pines about 1.9dB and 7.4dB for the reconstructed peak signal to noise ratio,and the reconstruction time is shortened about 8.6s and 2.5s.2,Combined the idea of different sampling rate every band image,a compressed sensing hyperspectral image reconstruction algorithm based on variable sampling rate is proposed.Hyperspectral images is grouped in the data collection,sampling the reference image using a high sampling rate,and the non-reference image using a low sampling rate;when reconstructing the image,first,the measurements of the reference image and the non-reference image is reconstructed,and then the original value difference between the images is reconstructed,thereby recovering the non-reference image.According to the simulation results,at the variable sampling rate,the reconstructed peak signal noise to ratio of the image Indian pines and the image Pavia University is improved about 1.9dB and 2.5dB compared with the constant sampling rate;and shortened about 2s and 2.8s.The variable sampling rate is superior to the constant sampling rate in both reconstruction time and reconstruction effect.