Also, due to the uncertainties in the environmental parameters, they risk not detecting plumes that are distorted by atmospheric and other environmental uncertainties. These approaches have the ability to simultaneously detect the plume and potentially identify the plume’s chemical constituents, which is a key difference between them and the methods we describe next which only detect plumes.Methods that don’t use a chemical spectral library are based on a statistical or data analytical transformation applied to the data. These include principle components, independent components, entropy, Fourier transform, and several other combinations or modifications, e.g. see [4]. These methods do not explicitly take advantage of the signal formulation physics, and therefore don’t exploit all available information in the data.
They also risk producing features/artifacts that have no obvious physics-based interpretation. Finally, they also rely on an analyst to recognize ��plume-like�� objects and distinguish them from non-plume features.In this paper we introduce a plume detection method that avoids some short-comings of both previously mentioned methods but also has features in common with both. This new method is not intended to replace current methods but rather to complement them. It is physics-based but it is not defined by the members of any specific collection of chemicals, large or small. Instead it uses surrogate chemical spectra which form a basis set for the set of all possible chemical spectra. The method has been applied to both real and synthetic hyperspectral imagery.
Only the results from synthetic data are presented here but results on real datacubes are similar. Section 2 presents the physics-based model. Section 3 presents matched filter detection and the basis vector method. Section 4 presents experimental results on a synthetic HSI datacube and conclusions are presented in Section 5.2.?Physics-based Radiance ModelIn this section we present the Drug_discovery three-layer physics-based radiance model which describes the basic physics of radiative transfer in the context of plume detection [1, 3, 5]. We present the model as a function of wavelength, �� (in ��m).This model can be written as:Lobs(��)=��a(��)[(1�\��p(��))B(Tp;��)+��p(��)Lg(��)]+Lu(��)+n(��)(1)where Lobs(��) represents sensor-recorded radiance in W/(m2 * sr * ��m) at wavelength �� (��m), ��a(��) and ��p(��) are dimensionless terms representing the atmosphere and plume transmissivity, respectively, B(Tp;��) has radiance units and is Planck’s Blackbody function at wavelength �� and plume temperature Tp (K), Lg(��) and Lu(��) are the ground-leaving and atmospheric upwelling radiances, respectively, and n(��) includes unmodeled effects and sensor noise [6].