Hyperspectral Data Processing: Algorithm Design and Analysis

22.6 Conclusions

The concept of DDA revolutionizes the traditional fixed-size band allocation in the sense that the former can vary band numbers according to various applications as opposed to the latter which uses a fixed number of bands for BS. A similar idea was also previously explored for a single hyperspectral signature in hyperspectral data Wang and Chang (2007). Its idea is derived from variable-length coding widely used in source coding where different source alphabets require different coding lengths for their own code words in accordance with their occurrence frequencies. This same idea can be applied to hyperspectral band selection provided that each source alphabet and its coding length are interpreted as a hyperspectral signature such as an endmember and the number of bands used to characterize the signature. Within this context various hyperspectral signatures will require their own variable numbers of bands as well as different bands to better describe their spectral characteristics (see Chapter 27). This is particularly true for LSMA where the signatures used to form a signature matrix to perform spectral unmixing certainly exhibit different spectral profiles for their own characterization. As a result, each signature used by LSMA has a different degree of difficulty with discrimination for its own identity and thus, requires a different set of bands to characterize its own spectral profile. The development of DDA arises from this need and allows users to allocate ...