Hyperspectral Data Processing: Algorithm Design and Analysis

20.5 Hyperspectral Compression by PSDP

PSDP provides a backbone for two major dual processes, progressive spectral dimensionality reduction (PSDR) and progressive spectral dimensionality expansion (PSDE), that can be used as a data dimensionality process for hyperspectral information compression.

20.5.1 Progressive Spectral Dimensionality Reduction

PSDR is a process that allows users to reduce a number of spectral components gradually by removing one spectral component at a time with decreased spectral component dimensionality priorities. It starts with a maximal number of spectral components and begins to reduce a small number of spectral components at a time by eliminating spectral components with lowest priorities until performance of data processing is not satisfied or it reaches the minimal number of spectral components that can be determined by VD. By implementing PSDR more and more hyperspectral information compression can be achieved by gradually reducing spectral information with removal of additional spectral components from the set of spectral components currently being considered. In what follows, we describe its implementation in detail.

PSDR

1. Prioritize all the transformed components via a DP criterion.

2. Initialization: Use VD to determine the minimal number of transformed components required to be retained, denoted by n_VD. Let n_initial be the number of spectral components for the process to begin with and n_Δ be the step size of dimensions to reduce. Set k