A real-time implementation of the method of principal components applied to dual-polarized radar returns

Summary form only given. Experiments performed with dual-polarized Ku-band radar systems have shown that there are distinct differences between the information contained in the like- and cross-polarized returns from the ice floes, particularly between those returns from new and old ice. In order to present the two different images on one monochrome display, it is necessary to combine them. The process can be expedited by using singular-value decomposition (SVD) to determine the eigenvectors, since, in doing so, it is not necessary to compute the covariance matrix explicitly. For the special case of transforming two input images into one output image, the SVD can be computed in a straightforward manner using the rotation matrix of Hestenes (1958). By performing the image transformation using parallel processors, an efficient pipelined architecture for computing the method of principal components can be realized. Such an architecture has been simulated on the Warp systolic computer and applied to the like- and cross-polarized radar images.<>