This chapter describes small-target detection of hyperspectral images using
remote sensing product-deriving techniques. The work reported in this chapter
is the further development of the work described in Chapter 11 for the purpose
of assessing and validating the effectiveness of SNR-increasing techniques.
The goal of the assessment is to examine how a hyperspectral datacube after
denoising (or after increasing the SNR using the techniques in Chapter 11) can
help improve the accuracy of the derived products or increase the confidence
of the remote sensing applications. Target detection is a hyperspectral image
application that is used in this chapter as another example to assess the hybrid
spectral-spatial noise reduction (HSSNR) technique.
Target detection from hyperspectral imagery has been an active research
area since the 1990s. Hyperspectral target-detection algorithms can be basically
classified into two categories: spectral-only and spectral-spatial. The spectral-only
algorithms use the known spectral signatures of the targets; this category
of algorithms includes the spectral matched filter, spectral angle mapper
(SAM), and linear mixture models. The spatial-spectral algorithms apply
when there is no known target spectral signature. They locate pixels that
display different spatial and/or spectral characteristics from their surroundings.
The spatial-spectral algorithms can be further divided into local- and