Sparse Unmixing Method Of Hyperspectral Image Based On Low-rank Priori

Hyperspectral remote sensing technology has developed rapidly since the 1980s.Due to its characteristics of“image-spectrum merging”and high spectral resolution,it has been widely used in many fields,such as agricultural production,environmental monitoring,urban planning and military exploration,and plays an increasingly important role.However,due to the complexity of ground object distribution and the imperfection of spectral acquisition technology,the spatial resolution of images obtained by hyperspectral remote sensing is low.Hence,there are many kinds of ground objects in one pixel,which is called mixed pixel.In order to improve the accuracy of the subsequent remote sensing image analysis,it is very important to unmix the hyperspectral image,that is,to extract the pure ground object spectrum contained in the image and estimate the proportion of these ground objects in the mixed pixels.In order to improve the unmixing accuracy of hyperspectral image,this paper makes full use of the spectral correlation and the spatial information between pixels of hyperspectral images under the sparse unmixing framework.The main contributions of this paper are as follows:?1?A sparse unmixing method for hyperspectral images based on nonlocal nuclear norm constraint is proposed.This method introduces a nonlocal nuclear norm constraint regularization term to take advantage of the nonlocal self-similarity of hyperspectral images.The nuclear norm is used to effectively constrain the abundance matrix of nonlocal blocks and promote it to keep low-rank.This method also introduces the collaborative sparse and the total variation?TV?term to consider the spectral correlation and the spatial information between adjacent pixels,respectively.?2?A sparse unmixing via superpixel based collaborative sparse and L_1/2 low rank prior for hyperspectral images method is proposed.The local homogeneous region of hyperspectral image cannot be well obtained by using square local blocks.Therefore,this method employs superpixel segmentation method.Then,the collaborative sparse and L_1/2 norm are utilized for each superpixel to consider its spectral correlation and spatial information.The method also introduces the TV term to consider the local spatial correlation between adjacent pixels.?3?A weighted nonlocal low-rank tensor decomposition method for sparse unmixing of hyperspectral images method is proposed.This method utilizes the tensor decomposition method to unmixing,which better maintains the structure of the hyperspectral image.In addition,this method uses the nonlocal self-similarity of hyperspectral image to maintain low-rank property of abundance image.This method also combines collaborative sparse and TV terms for unmixing.The methods proposed in this paper are compared with several state-of-the-art methods on simulated datasets and real hyperspectral datasets.The experimental results show the superiority of the proposed methods.