Hyperspectral Data Processing: Algorithm Design and Analysis

19.4 Progressive Spectral/Spatial Compression

The spectral/spatial compression presented in Section 19.3 requires the knowledge of q that is the number of dimensions needed to be retained for DR and p that is the number of bands needed to be selected for BS to achieve spectral compression. Despite that VD can be used to estimate the values of the q and p, both the q and p in fact vary with applications as discussed in Chapter 5. In order to avoid dealing with the issue of determining the q and p, an alternative approach called progressive spectral compression for DRT and DRBS will be developed in Chapters 20–21 to perform DRT and BS progressively via dimensionality prioritization and band prioritization, respectively. Its idea can be depicted in Figure 19.10(a) and (b), which is very similar to Figures 19.5–19.9 except DRT/DRBS replaced by progressive DRT/progressive DRBS (Figure 19.11).

Figure 19.11 Progressive spectral/spatial compression.