Hyperspectral Data Processing: Algorithm Design and Analysis

21

Progressive Band Dimensionality Process

Using DR as a means to perform progressive spectral dimensionality process (PSDP) by dimensionality prioritization (DP) has been studied in great detail in Chapter 20. It is natural to include Chapter 21 as a companion chapter of Chapter 20 with its main theme focused on band dimensionality for progressive process. In other words, what has been done for PSDP via DP can also be applied to progressive band dimensionality process (PBDP) via band prioritization (BP). However, there are a few significant differences between PSDP and PBDP that make the techniques developed for PSDP in Chapter 20 not necessarily applicable to PBDP. One is how data information is represented by DP and BP. In PSDP the data information is compacted via a transformation from the original data space to a new transformed data space compared to PBDP that only retains the information of selected bands but completely discards the information of unselected bands. So, PSDP requires the complete hyperspectral image cubes to perform transformation as opposed to PBDP that only processes the hyperspectral image cubes formed by selected bands. Another major difference is that PSDP requires a pair of parameters, projection vectors to specify transformed components in the transformed data space and priority scores to prioritize the information contained in the transformed components, whereas PBDP does not require projection vectors but only needs priority scores to prioritize ...