VIII
Applications
As a final part to conclude this book a few applications are of particular interest to form Category C. Chapter 30 includes two applications of hyperspectral target detection, subpixel target size estimation and concealed target detection, both of which have unique issues to be addressed and cannot be resolved by spatial domain-based techniques due to the fact that the targets of interest are invisible either completely or partially to human eyes by inspection. Under such circumstances detection of these targets must rely on their spectral characteristics rather than on their spatial properties. These applications provide good examples to demonstrate unique advantages of hyperspectral imaging over traditional image processing. On the other hand, as noted in Chapter 1 (Sections 1.2 and 1.3), hyperspectral imagery shall not be considered as a straightforward extension to multispectral imagery by simply adding more spectral bands. As a matter of fact, it is spectral resolution, not total number of spectral bands, that matters. Consequently, techniques developed for hyperspectral imaging are not necessarily applicable to multispectral imaging, specifically linear spectral mixture analysis (LSMA). Chapter 31 presents two nonlinear band dimensionality expansion techniques, band expansion process (BEP), and kernel trick for kernelization, to make hyperspectral imaging techniques also effective on multispectral imagery. Since hyperspectral imaging techniques generally ...
