In this paper, we investigate the use of airborne radar images to detect localised objects in a background. Methods of processing the Synthetic Aperture Radar (SAR) data are discussed, with descriptions of programs to implement these methods given. These methods include the use of a Kruskal-Wallis test and the construction of a Probability Density matrix.
The aim of these investigations is to develop methods based on pixel group characteristics that will discriminate between the true and the false targets, allowing as many false targets to be eliminated without removing too many true targets.
The data that is examined in the research is described in both a qualitative and a quantitative manner. Fifteen combinations of the methods are studied. Two of these methods were successful in eliminating many false targets while detecting most true targets. One method involved the construction of a Probability Density Matrix with a 3 x 3 window size, followed by the Kruskal-Wallis test with a 4 x 4 window size. The other method involved the construction of four Probability Density Matrices with 2 x 1 window sizes. Then combining the results by multiplication of the pixels. This was subjected to a 4 x 4 Kruskal-Wallis test.
The paper recommends investigations into combining the results of 2 x 1 probability mappings using addition of pixels or constructing a Probability Density Matrix from the four 2 x 1 results using a 2 x 2 window size.