Hyperspectral Data Processing: Algorithm Design and Analysis

4.2 Simulation of Targets of Interest

As noted earlier, a target used in this book is referred to as an object whose existence can be spectrally characterized by certain properties, for example, statistics of spectral correlation across the wavelength range used for data acquisition. Generally two types of targets are of interest in hyperspectral target analysis, subsample targets, and mixed-sample targets, as discussed in Chapter 2.

4.2.1 Simulation of Synthetic Subsample Targets

First, we simulate a subsample target. Assume that a subsample target is specified by a signature, p, for example, panel signature in Figures 1.8–1.10, 1.12(c), (d) and 1.16–three subsample targets with ¾, ½, and ¼ sample sizes, respectively. Figure 4.1(a) shows how a subsample target t₁ with ¾ sample size is simulated by the panel signature, p. To simulate the subsample target t₁, we first simulate one-sample vector specified by a background signature b and three-sample vectors specified by p to form a four-sample square panel as shown at the bottom layer of Figure 4.1(a).

Figure 4.1 Simulations of three subsample target panels: (a) subsample target panel, p₁ with ¾ of p + ¼ of b; (b) subsample target panel, p₂ with 1/2 of p + 1/2 of b; (c) subsample target panel p₃ with ¼ of p + ¾ of b.

The four-sample square panel is then shrunk to its ¼ size by averaging all four sample vectors to a single four-subsample ...