Abstract

Synthetic Aperture Radar (SAR) data provides very high-resolution images under every kind of weather condition. It penetrates through darkness, clouds, rain etc., and is very useful in detecting after-effects of disasters like earthquakes, landslides, floods etc. The present study proposes data fusion of SAR data with Multispectral data using histogram based Higher Order Singular Value Decomposition (HOSVD).
The input images to HOSVD data fusion process are radiometrically corrected, precession rectified SAR and multi-spectral imagery. The author uses histograms corresponding to the intensity (pixel) values of SAR and Multispectral (MS) data. Histogram bins are generated for SAR data, with each bin representing a certain intensity value range. In other words, each bin is now
represented by its frequency i.e., the number of pixels within the range of the bin. In this illustration, image pixel intensities were classified into eleven bins, each with a range of 25 intensities. Eleven variance-covariance matrices corresponding to multispectral data of each bin associated with SAR data are generated. These variance-covariance matrices (or their correlation matrices) are subjected to SVD analysis. Data fusion is performed for each of these binned data as a two-step procedure. A rotation about SVD-eigen axis is performed and the component with largest SVD is replaced with corresponding SAR pixel data. An inverse rotation is subsequently applied, to return to the original axis, thereby, generating a data fused image. The data from these eleven fused images are added, to realize the final data-fused image.
The HOSVD data fusion process is illustrated with PAN and MSS Images of Quickbird 2 satellite at 60 cm and 2.4 meters. Data fused images of SAR-multispectral highlight submerged areas after flooding. An analysis of one of the correlation matrices using SVD, summarizes the improvement to radiometric fidelity.