Spatial Resolution Enhancement For CHANG’E-1Hyperspectral Imagery

To broaden the future application of CHANG’E-1imagery, includinghyperspectral imagery (low spatial resolution of200m) and CCD imagery(relatively high spatial resolution of120m), this dissertation has carried on asystematic research on resolution enhancement techniques for CE-1imagery.This research aims at improving the spatial resolution while preserving thespectral information as much as possible, thus the enhanced hyperspectralimagery will be able to meet the demands of future lunar terrain exploration.This dissertation divided its research into two parts: resolution en hancementbased on image fusion and resolution enhancement without an auxiliary image.In the category of resolution enhancement technique based on image fusion, thisdissertation proposed two different methods: ARSIS/PCNN model andNLPCA/INDUSION technique. The ARSIS/PCNN model focuses on searchingand modeling a relationship between the high frequencies of the images to befused for missing information, and it preserves the spectral content of theoriginal image for ARSIS concept’s very definition. While theNLPCA/INDUSION technique focuses on data compression/reduction and detailenhancement, which leads to smoother detail information and higher spatialresolution. In the category of resolution enhancement technique without anauxiliary image, this dissertation also proposed two different methods:spatial-spectral super-resolution and super-resolution via sparse representation.The spatial-spectral super-resolution pays more attention to extracting edgepixels, spectral unmixing and sub-pixel mapping, with distinctive features suchas short elapsed time and high spectral fidelity. While the super-resolutiontechnique via sparse representation focuses on training redundant dictionariesand establishing the mapping relation between the high-resolution image patchesand low-resolution image patches. The insensitivity to training images broadensthe applicability of the algorithm to a great deal.Furthermore, after thorough qualitative and quantitative analysis on theexperimental results, it is demonstrated that the visual improvement and spectral fidelity of the proposed methods outperform many conventional methods ofresolution enhancement.