Spectral library material separability using WorldView-3 and Landsat-8 spectral bands

Abstract

The WorldView-3 and Landsat-8 satellites are the most recently deployed systems in their constellations and the unique data from these sensors can positively impact environmental and military target detection applications. The research team uses spectral library data in the VNIR and SWIR spectral bands of WorldView-3 and Landsat-8 to determine the best combination of spectral bands and spectral distance measure to yield the largest spectral distance value for each target material. Spectral distance measures include Euclidean Distance, Spectral Angle Mapper, Spectral Correlation Measure, and Spectral Information Divergence. The optimal configuration results are stored in a look-up-table for implementation in an automated target detection system. The Freedman-Diaconis and Shimazaki-Shinomoto methods for optimal histogram bin width determination are applied to spectral distance measures that are cross computed for each material in the spectral library and for each sensor. The bin width determination is used to characterize material clusters based on intercluster and intracluster spectral distances. The material cluster characterization results are stored in a look-up-table for fast histogram based initialization of clustering algorithms. The research team uses the in-band spectral library data for determining end member abundance estimates based on combinations of spectral bands, end member combinations, spectral distance measure, and additive white Gaussian noise for both sensors. The endmember abundance estimates are optimized using Differential Evolution, Least Squares, and Linear Simplex. The numerical accuracy of the end member abundance determination is compared across the three optimization algorithms. The completion of this foundational work increases the data exploitation potential of WorldView-3 and Landsat-8 by providing a fundamental characterization of material separability with respect to these sensors.