A Fusion Method for Mixed Pixels based on Classification of Panchromatic Image

A low-spatial-resolution multispectral (MS) mixed pixel superimposes multiple high-spatial-resolution pixels, some of which are mixed and others are pure. The fused versions of MS mixed sub-pixels normally remain spectrally mixed and visually blurred. A high-resolution classification map is a requisite to the proper fusion of MS mixed sub-pixels. In this article, soil sub-pixels and mixed sub-pixels within a MS image are initially separated to delineate coarse boundaries between vegetation and non-vegetation. Then, the high-spatial-resolution panchromatic (PAN) pixels near such boundaries will be classified into vegetation and non-vegetation, regarding to the spatial relationships among objects and related shadows. Finally, each MS mixed sub-pixel with a known class will be set to be a pure pixel in a fusion process, and others will be fused as pure sub-pixels. The fusion method proposed in this paper offers synthetic products with high spectral accuracy and clear soil-vegetation boundaries in a trial upon IKONOS MS and PAN images.