Hyperspectral Data Processing: Algorithm Design and Analysis

21.2 Band Prioritization

The concept of BP was implicitly used by Chang et al. (1999) in conjunction with information divergence-based BD to perform BS. Its potential in various applications has not been realized since then. This section revisits BP and extends it to PBDP while postponing progressive BS (PBS) to Chapter 23, which can be considered as an application of PSDP to BS.

For each spectral band B_l BP prioritizes B_l according to its contained information measured by a custom-designed information criterion. PBDP selects bands progressively by expanding or reducing band dimensionality. Such a progressive band dimensionality process can be terminated by a stopping rule that is determined by various applications instead of a specific value of p that must be selected by BS in advance. Nevertheless, in this case, VD can be used to provide a reasonable lower bound and a reliable upper bound on the value of the . In addition to the second-order statistics proposed in Chang et al. (1999) this chapter extends criteria to high-order statistics. Specifically, four categories of criteria are considered in this chapter for BP. The first category is made up of second-order statistics-based criteria, which include second-order statistics-based variance and signal-to-noise ratio (SNR). The second category contains higher-order statistics-based criteria including skewness, kurtosis, entropy, ...