Hyperspectral And Multispectral Image Fusion Based On Local Low Rank And Spectral Unmixing

A spectral image is a set of 2-D images(that can be seen as a 3-D cube)representing the reflectance or radiance of a scene in different parts of the electromagnetic spectrum.In the field of remote sensing,these play an important role in terrain classification,mineral detection and exploration,environmental monitoring,military surveillance,and space observation.There are two typical spectral images,hyperspectral images(HSIs)and multispectral images(MSIs),that are widely used in remote sensing for earth observation systems and regional applications.A HS imaging system provides detailed spectral information that enables the observer to detect and classify a pixel based on its spectral characteristics.However,the spatial resolution of HSI systems is generally lower than those of MSI systems.An MS imaging system has higher spatial resolution but fewer spectral channels than HSI,which leads to lower spectral resolution.In several practical applications,the low spatial resolution of HSI results in a large number of mixed pixels,and significantly degrades analysis capabilities in comparison with requirements in the civil and military fields.Hence,reconstructing a high spatial resolution HSI(HHSI)from two degraded and complementary images is a challenging but crucial issue.In this paper,we propose a fusion algorithm by combining linear spectral unmixing with the local low-rank property.By taking advantage of the local low-rank property,we first partition the corresponding spectral image into patches.For each patch pair,we cast the fusion problem as a coupled spectral unmixing problem that extracts the abundance and the endmembers of MSI and HSI,respectively.It then updates the abundance and the endmember through an alternating update algorithm.In fact,the convergence of the alternative update algorithm can be mathematically and empirically supported.We also propose a multiscale postprocessing procedure to combine fusion results obtained under different patch sizes.In experiments on three data sets,the proposed fusion algorithms outperformed state-of-the-art fusion algorithms in both spatial and spectral domains.