### Table 4. Characteristics of Angular Signatures in Spectral Space

"... In PAGE 15: ... The angular signatures are nearly linear in the spectral space. In fact, statistical analysis indicates a significant linear relation between the red and near-infrared reflectances ( Table4 ). Canopy reflectances having the same normalized difference vegetation index (NDVI) value lie on a single line passing through the origin of the red-NIR plane.... In PAGE 16: ...he second part of this series (Zhang et al., 2001; also see Kaufmann et al., 2000). The view direction averaged NDVI values of the different biomes, shown in Table4 , indicate the unique inclination of the angular signatures in the spectral space, with the exception of grasses and crops. The location of the biome data in the spectral space is also distinct as it can be ascertained from the mean red and near-infrared reflectance values shown in Table 4.... In PAGE 16: ...he second part of this series (Zhang et al., 2001; also see Kaufmann et al., 2000). The view direction averaged NDVI values of the different biomes, shown in Table 4, indicate the unique inclination of the angular signatures in the spectral space, with the exception of grasses and crops. The location of the biome data in the spectral space is also distinct as it can be ascertained from the mean red and near-infrared reflectance values shown in Table4 . A methodology for quantifying the three metrics characterizing the angular signatures in spectral space is given in the next section.... In PAGE 19: ... For the pixel- mean signatures shown in Fig. 8, the slope and intercept values are given in Table4 . The forest biomes show a larger slope, consistent with detailed pixel level calculations presented in Table 5.... In PAGE 31: ... Distribution of Biomes Based on Homogeneity Factors Table 3. Geometrical Characterization of POLDER Data Table4 . Characteristics of Angular Signatures in Spectral Space Table 5.... ..."

### TABLE III Modification of spectral signatures to incorporate spectral confusion. We show the spectral regions in which there is spectral overlapping and the non-modified and modified classes. Let us explain the process we followed to create the new spectral signatures in greater detail. With n being a non-

### TABLE 7 Actual and predicted performance of the SDC algorithm with Newton iteration for the spectral decomposition along the pure imaginary axis.

1997

Cited by 32

### Table 1. Examples of mass spectral signature compounds for different types of threat agents.

in W. A. Bryden

"... In PAGE 4: ... The spectra of these smaller fragments are then fully analyzed and refined to produce verifiable signatures of the substances of interest. Examples of categories of signature com- pounds for many different types of threat agents are shown in Table1 . It is evident that the mass spectr has the capability to be the universal agent detector for counterproliferation applications once the portability 1 and is able 2).... ..."

### Table 1: Values used for the three class-dependent parameters for simulating 6 spectral signatures.

### TABLE II Values used for the three class-dependent parameters for simulating 6 spectral signatures.

### Table5.5forestimatesofclassproportions. .................. 75 5.9 ClassBasisSpectrafortheFirstSetofSpectralSignatures ......... 80 5.10EstimatesoftheClassSpectralSignatures................... 81 5.11 Modified Class Basis Spectra for the First Set of Spectral Signatures . . . . 83 5.12EstimatesoftheClassSpectralSignatures................... 84

### Table 3. Comparison: Results by feature for a purely spectral Fisher discriminant and intelligent threshold for our training scene and test scene.

in A Genetic Algorithm for Combining New and Existing Image Processing Tools for Multispectral Imagery

"... In PAGE 9: ...ame features in the images shown/described above. This approach is based purely on spectral information. On application to the data used in the training run (Fig. 4), and on application to the out-of-training-sample scene, this traditional approach produced results summarized in Table3 . On the training scene, the Fisher discriminant is in each case able to find a satisfactory match to the training data, though still below the performance of the algorithms produced by the Genie system.... ..."

### Table 8 Condition numbers ( 2) for the Jacobian (J = Dxf), the corresponding companion matrix, and the spectral condition number, s , associated with the pure imaginary eigenvalues. Pt

1996

"... In PAGE 16: ... We expect that, for small, the perturbation of the eigenvalues will be less than = jjBjj2=s( ) and we consider the eigenproblem well-conditioned if s( ) is near one. The rst column of Table8 presents the spectral condition number for the critical eigenvalues of the Jacobian at each selected point solution. Clearly, the data show that methods based on explicit determination of the spectrum (JGR) or their sums (BP) directly from the Jacobian entries should be relatively insensitive to small perturbations to the elements of J.... In PAGE 16: ... Table8 also shows 2-norm condition estimates for the Jacobian matrix and its associated companion matrix as computed by Algorithm 1 at the selected points; in each case the reduction results in an in ation of the condition number, but the increase is mild (at most three orders of magnitude). Thus, the main source of instability in the computation of the characteristic polynomial coe cients { the use of non-unitary similarity transformations in the reduction of the Hessenberg form { is well-behaved along the branch of Hopf points.... ..."

Cited by 4