Research On 3D Discrete Cosine Transform Dictionary Denoising Algorithm For Hyperspectral Data

With the development of hyperspectral technology,hyperspectral data is widely used in many fields,such as agricultural remote sensing,military reconnaissance and geological exploration.Due to the high spatial resolution and spectral resolution of hyperspectral data,the light signal that can enter the sensor is very weak.And the acquisition device is easily affected by the calibration error,so that the acquired hyperspectral image contains a lot of noise.It not only affects the visual effect of the spatial domain,but also makes the spectral domain signal distort,and affects the spectral feature analysis precision and information extraction.Therefore,it is significant to study a denoising algorithm that can maintain spatial domain image details and effectively remove noise from spatial and spectral domain.In order to solve the problems that noise intensity of each band for hyperspectral data is different and noise exists in both spatial and spatial and spectral domains.In this paper,the classical denoising algorithm is applied to denoise the simulated hyperspectral and real tea data respectively from spectral denoising algorithm in spectral domain and band by band denoising algorithm in spatial domain.Based on that,three dimensional fast Fourier transform,grouped three dimensional discrete cosine transform dictionary and block matching 3D discrete cosine transform dictionary denoising algorithm based on guided filter are proposed.Because of the existence of Gauss random noise in the whole frequency domain,the noise reduction effect of three-dimensional FFT in space domain is poor.Compared with classical denoising algorithm,Grouped 3D DCT dictionary and Guided block matching 3D DCT dictionary improve the signal-to-noise ratio in the spatial and spectral domain,and keep the edge and detail of the image well.The spectral characteristics and noise types of hyperspectral data are analyzed.The application of sparse representation denoising theory of 3D FFT and 3D DCT dictionary in 3D hyperspectral data is mainly studied.The specific contents are as follows:(1)First,the spectral characteristics and spectral correlation of hyperspectral data are analyzed,providing the basis for grouping in hyperspectral bands.Then,the local mean standard deviation method is used to estimate the noise standard deviation of hyperspectral data,providing a reference threshold for denoising algorithm.(2)Three dimensional fast Fourier transform denoising algorithm is studied.Fast Fourier transform denoising is performed in the spatial and spectral regions of the hyperspectral data,respectively.The spatial domain and the spectral domain noise are suppressed.(3)Grouped three dimensional discrete cosine transform dictionary denoising algorithm is studied.The hyperspectral images are grouped according to the correlation degree between the spectra.Three dimensional DCT transform is carried out on the 3D data after grouping,and the Orthogonal Matching Pursuit algorithm is used to recover the clean image information in 3D DCT domain.(4)Guided block matching 3D DCT dictionary denoising algorithm is proposed.The algorithm combines the non local mean search similar block,grouped three dimensional DCT dictionary and guided filter.The denoising process makes full use of the high correlation between block and space redundancy and the image details of the best signal to noise ratio,and the denoising effect is further improved.The experimental results show that for the simulated hyperspectral data and the real tea data,the grouping DCT 3D dictionary is superior to the classical denoising algorithm in the spectral domain,and the denoising effect is obvious.After improvement,the guided block matching three dimensional DCT dictionary denoising algorithm not only has the highest signal-to-noise ratio in the spectral domain,but also keeps the best detail in the spatial domain,which is the best algorithm of denoising effect in this paper.