Hyperspectral Data Processing: Algorithm Design and Analysis

23.3 Progressive Band Selection

The PBDP (Chapter 21) allows users to select bands progressively back and forth for band dimensionality reduction and expansion at discretion without actually determining the number of bands needed to be selected. In the traditional BS, the value of this p must be first determined and then fixed at a constant value during the whole data processing. When the value of is not desired and a new value is needed, the entire BS process must be reimplemented where the previous results obtained for old values of cannot be reused for the new value of because new bands will be selected by solving a new set of BS optimization problem. This type of BS is referred to as fixed-dimensionality BS (FDBS) to emphasize the role of p to be fixed at a constant value instead of using static band selection (SBS), discussed in Chapter 22. Most importantly, such FDBS may not be applicable or effective in hyperspectral data exploitation. For example, as discussed in Chapter 22, various material substances have their own spectral characteristics in signature profiles that may present ...