Hyperspectral Image Denoising Method Based On The Nonstationary Representation Model

At present,hyperspectral remote sensing technology is developing rapidly and widely used.In order to obtain high quality image data,the image quality can be improved by means of image processing when the equipment is limited.However,the hyperspectral imagealways have been contaminated by the conditions while acquiringand then denoising is anindispensable step in the image processing.So the hyperspectral image denoising method based on the nonstationary representation model is proposed in this paper.In the hyperspectral image,from the spectral dimension,a single pixel has a complete spectral shape and can be used to determine the corresponding elements of the surface.From the spatial dimension,itobtains the position information on the ground and the arrangement with other pixels in the position of the spatial,which is the characteristic of the "spatial-spectral" of the hyperspectral image.And because of the low spatial resolution of the sensor,the range of the ground corresponding to each pixel is wide and contains more features.The spectral signatures of the pixel is a mixture of many kinds of materials.In order to solve this problem,the spectral unmixing of hyperspectral image refers to finding the number of endmembers and their abundance and then thedenoised hyperspectral image is derived by image reconstruction.This is based on the spectral unmixingmethod for the denoising.The non-stationary representation method proposed in this paper is based on the denoising method of the spectral decomposition.The main contents are as follows:1?According to the linear spectral mixture model,these pixels and their weights are used in the eigen-representation method,and the unmixing of the central pixels.The "real" endmember and their abundance Value for the image reconstruction to get the recovery of the hyperspectral image,that is,the image after the denoising.2?There is spatial nonstationarity for hyperspectral image.Based on the non-stationary modeling method proposed by Fuentes et al.,The non-local mean method is used to find the pixels with similarity to the central pixel,using the Mahalanobis distance and Euclidean distance respectively calculate the similarity between image blocks as weight.3?The experiments were carried out by using simulated hyperspectral image and real hyperspectral image.Meanwhile,the original image is preliminarily and quantitatively evaluated,and the denoised image is same as before.The peak signal to noise ratio(PSNR),structural similarity(SSIM),root mean square error(RMSE)of simulated image and the SNR of real image are calculated respectively.The experimental results show that the model can denoise the hyperspectral image at the same time in the spectral dimension and spatial dimension.Compared with other methods,the method proposed in this paper has good robustness and can keep the more spatial texture information of the image.