Hyperspectral Data Processing: Algorithm Design and Analysis

22.5 Experiments for Dynamic Dimensionality Allocation

As demonstrated in PSDP and PBDP different signatures required various spectral dimensions/bands to effectively perform linear spectral unmixing due to their different spectral characteristics. These experiments provided evidence that a signature should have its own set of spectral dimensions/bands to characterize its spectral profile. In order for PSDP and PBDP to demonstrate progressive performance, three applications, endmember extraction, land cover/use classification, and spectral unmixing, were considered in Chapters 20 and 21 as the parameters, q and , varied ranging from n_VD to 2n_VD without knowing exactly what the values of q and were. DDA presented in Section 22.2 is developed to make an attempt of narrowing down the range of [n_VD,2n_VD] to best possible values that can be specified by DDA for dimensionality allocations for various signatures. In the following sections we calculate DDA for the three different image data sets which have been used for experiments conducted in Chapters 20 and 21 where the reference signatures were chosen to be the signature means of signatures of interest. Of course selection of an appropriate reference signature is also an interesting issue which will be discussed in Chapters 25, 27 and ...