Performance of spatial sharpening methods for hyperspectral imagery

摘要

"Pan" or single broadband sharpening of multispectral (low spatial resolution) imagery is currently deployed on airborne and satellite systems. The challenges for spatial sharpening of hyperspectral imagery are the focus of the current study, which utilizes high spatial resolution, geo-referenced multispectral imagery available from the QuickBird satellite with low spatial resolution AVIRIS hyperspectral imagery. Performance analysis of a spectral normalization method known as the CN Spectral Sharpening (CNSS) enables correction for the mismatch in spectral radiance levels of the two input images due to differences of sensor platform altitude, date of imaging, atmospheric path and solar irradiance conditions. The BAE Systems Spectral Similarity Scale is utilized to optimize the spectral match between the unsharpened input and sharpened output. Performance evaluation includes comparison of the histogram spectral means and standard deviations of selected regions of interest, combined with computing the spectral correlation difference matrix between the unsharpened and sharpened AVIRIS data. Significant similarity is demonstrated with high spectral correlation, yet high variance change between the green and red MSI channels results in a discontinuity region of the corresponding HSI bands. Future systems incorporating collocated high spatial resolution MSI with lower resolution HSI will enable automated spatial sharpening with unproved spectral accuracy.