Research On Destriping Methods Of Hyperspectral Images

With the rise and rapid development of hyperspectral remote sensing technology, hyperspectral remote sensing images have been widely used into both military and civil areas by the governments, research institutions, and companies around the world. The reason is attributed to that hyperspectral remote sensing images often have much abundant spatial and spectral information.However, during the process of acquisition or transmission of the hyperspectral remote sensing images, they may be corrupted due to the impacts of various factors, such as atmosphere, the precision of sensors, and so on. The existence of noise may severely degrade the quality of the imagery and bring difficulties in subsequent process-ing. Therein striping noise is a typical example. Recently, the problem of destriping of hyperspectral images has attracted an increasing interest in remote sensing community. A variety of methods have been proposed to effectively alleviate the effects of the strip-ing noise. Nevertheless, most of them fail to take full advantage of the high spectral correlation between the observation subimages in distinct bands and consider the local manifold structure of the hyperspectral data space. In order to remedy this drawback, in this thesis various destriping algorithms and corresponding applications have been investigated. Moreover, a novel graph regularized low-rank representation destriping (GRLD) algorithm is proposed. The main work of this thesis is summarized as follows:(1) A variety of traditional and state-of-the-art destriping algorithms are investigated and discussed in depth. In addition, their advantages and disadvantages are also compared.(2) A novel graph regularized low-rank representation destriping (GRLD) algorithm is proposed, which not only take full advantages of the high spectral correlation between the observation subimages in distinct bands, but also can well preserve the intrinsic local structure of the original hyperspectral data. Consequently, the proposed method can effectively reducing striping noise while still well preserv-ing the useful information of the original data. Furthermore, the proposed method is compared with several state-of-the-art destriping methods. The experimental results and quantitative analysis demonstrate its effectiveness and superiority.(3) The proposed graph regularized low-rank representation (GLRR) method is an important extension of the original low-rank representation method. It is because that GLRR can both properly restore the potential subspace structure of the orig-inal hyperspectral data, and well preserve its local manifold structure.