Hyperspectral Data Processing: Algorithm Design and Analysis

27.6 Conclusions

This chapter presents a new approach to BS, called VNVBS for a single hyperspectral signature. Unlike most band selection techniques which are designed for images, VNVBS is designed for characterization of a single hyperspectral signature vector without a need of image sample correlation. To select appropriate bands, an OSP-BPC is also developed, which decomposes the original spectral signature vector into two orthogonal components that can be used for spectral characterization. Experimental results demonstrate that VNVBS is more effective in preserving information for hyperspectral signature characterization than that using full band information in hyperspectral signature characterization and analysis.