Mixed-pixel decomposition of SAR images based on single-pixel ICA with selective members

Abstract

In the decomposition of synthetic aperture radar (SAR) images with mixed pixels, two issues cannot be avoided; (1) there are more land cover types than the number of SAR image channels; and (2) the fixity of the land cover types impacts the precision of the decomposition. Based on the idea that varying the land cover type according to changes in pixels will lead to an improvement in decomposition accuracy, we propose the SM-SPICA (single-pixel Independent Component Analysis with selective members) algorithm that adjusts the land cover type dynamically. The algorithm is further based on mixed-pixel decomposition in a linear-spectrum mixed model and the relevant ICA theory. Experimental results using ENVISAT-ASAR images (both VV and VH) of Beijing confirm that SM-SPICA can decompose SAR images containing more land cover types than the number of image channels with greater accuracy.