Search

There have been a significant number of recent papers about hyperspectral imaging, which propose various methods for estimating the number of materials/endmembers in hyperspectral images. This is sometimes called the “intrinsic” dimension (ID) of the image. Estimation of the error variance in each spectral band is a critical first step in ID estimation. The estimated error variances can then be used to preprocess (e.g., whiten) the data, prior to ID estimation. A range of variance estimation methods have been advocated in the literature. We investigate the impact of five variance estimation methods (three using spatial information and two using spectral information) on five ID estimation methods, with the aid of four different, but semirealistic, sets of simulated hyperspectral images. Our findings are as follows: first, for all four sets, the two spectral variance estimation methods significantly outperform the three spatial methods; second, when used with the spectral variance estimation methods, two of the ID estimation methods (called random matrix theory and NWHFC) consistently outperform the other three ID estimation methods; third, the better spectral variance estimation method sometimes gives negative variance estimates; fourth, we introduce a simple correction that guarantees positivity; and fifth, we give a fast algorithm for its computation. [ABSTRACT FROM AUTHOR]

Many methods have been proposed in the literature for estimating the number of materials/endmembers in a hyperspectral image. This is sometimes called the “intrinsic” dimension (ID) of the image. A number of recent papers have proposed ID estimation methods based on various aspects of random matrix theory (RMT), under the assumption that the errors are uncorrelated, but with possibly unequal variances. A recent paper, which reviewed a number of the better known methods (including one RMT-based method), has shown that they are all biased, especially when the true ID is greater than about 20 or 30, even when the error structure is known. I introduce two RMT-based estimators ( R M T G , which is new, and R M T K N , which is a modification of an existing estimator), which are approximately unbiased when the error variances are known. However, they are biased when the error variance is unknown and needs to be estimated. This bias increases as ID increases. I show how this bias can be reduced. The results use semi-realistic simulations based on three real hyperspectral scenes. Despite this, when applied to the real scenes, R M T G and R M T K N are larger than expected. Possible reasons for this are discussed, including the presence of errors which are either deterministic, spectrally and/or spatially correlated, or signal-dependent. Possible future research into ID estimation in the presence of such errors is outlined.

Many papers in the hyperspectral literature use simulations (based on a linear mixture model) to test algorithms, which either estimate the “intrinsic” dimensionality (ID) of the data or endmembers. Usually, these simulations use “real-world” endmembers, proportions distributed according to a uniform or Dirichlet distribution on the endmember simplex, and Gaussian errors which are “spectrally” and “spatially” uncorrelated. When the error standard deviations (SDs) in different bands are assumed to be unequal, they are usually estimated using Roger's method. The simulated and real-world data in these papers are so different that one cannot be confident that the various advocated methods work well with real-world data. We propose a general methodology which produces more realistic simulations, providing us with greater insights into the strengths and weaknesses of various advocated methods. With the aid of the well-known Indian Pines and Cuprite scenes, we compare several specific options within the proposed methodological framework. We also compare the performance of five well-known ID estimators using both real and simulated datasets and demonstrate that Roger's SD estimates are positively biased. A proof that Roger's estimates are always positively biased is given. [ABSTRACT FROM PUBLISHER]

Determining the intrinsic dimension of a hyperspectral image is an important step in the spectral unmixing process and under- or overestimation of this number may lead to incorrect unmixing in unsupervised methods. In this paper, we discuss a new method for determining the intrinsic dimension using recent advances in random matrix theory. This method is entirely unsupervised, free from any user-determined parameters and allows spectrally correlated noise in the data. Robustness tests are run on synthetic data, to determine how the results were affected by noise levels, noise variability, noise approximation, and spectral characteristics of the endmembers. Success rates are determined for many different synthetic images, and the method is tested on two pairs of real images, namely a Cuprite scene taken from Airborne Visible InfraRed Imaging Spectrometer (AVIRIS) and SpecTIR sensors, and a Lunar Lakes scene taken from AVIRIS and Hyperion, with good results. [ABSTRACT FROM PUBLISHER]

Abstract: Tethered balloon remote sensing platforms can be used to study radiometric issues in terrestrial ecosystems by effectively bridging the spatial gap between measurements made on the ground and those acquired via airplane or satellite. In this study, the Short Wave Aerostat-Mounted Imager (SWAMI) tethered balloon-mounted platform was utilized to evaluate linear and nonlinear spectral mixture analysis (SMA) for a grassland-conifer forest ecotone during the summer of 2003. Hyperspectral measurement of a 74-m diameter ground instantaneous field of view (GIFOV) attained by the SWAMI was studied. Hyperspectral spectra of four common endmembers, bare soil, grass, tree, and shadow, were collected in situ, and images captured via video camera were interpreted into accurate areal ground cover fractions for evaluating the mixture models. The comparison between the SWAMI spectrum and the spectrum derived by combining in situ spectral data with video-derived areal fractions indicated that nonlinear effects occurred in the near infrared (NIR) region, while nonlinear influences were minimal in the visible region. The evaluation of hyperspectral and multispectral mixture models indicated that nonlinear mixture model-derived areal fractions were sensitive to the model input data, while the linear mixture model performed more stably. Areal fractions of bare soil were overestimated in all models due to the increased radiance of bare soil resulting from side scattering of NIR radiation by adjacent grass and trees. Unmixing errors occurred mainly due to multiple scattering as well as close endmember spectral correlation. In addition, though an apparent endmember assemblage could be derived using linear approaches to yield low residual error, the tree and shade endmember fractions calculated using this technique were erroneous and therefore separate treatment of endmembers subject to high amounts of multiple scattering (i.e. shadows and trees) must be done with caution. Including the short wave infrared (SWIR) region in the hyperspectral and multispectral endmember data significantly reduced the Pearson correlation coefficient values among endmember spectra. Therefore, combination of visible, NIR, and SWIR information is likely to further improve the utility of SMA in understanding ecosystem structure and function and may help narrow uncertainties when utilizing remotely sensed data to extrapolate trace glas flux measurements from the canopy scale to the landscape scale. [Copyright &y& Elsevier]

Many methods have been proposed in the literature for estimating the number of materials/endmembers in a hyperspectral image. This is sometimes called the "intrinsic" dimension (ID) of the image. A number of recent papers have proposed ID estimation methods based on various aspects of random matrix theory (RMT), under the assumption that the errors are uncorrelated, but with possibly unequal variances. A recent paper, which reviewed a number of the better known methods (including one RMT-based method), has shown that they are all biased, especially when the true ID is greater than about 20 or 30, even when the error structure is known. I introduce two RMT-based estimators ( R M T G , which is new, and R M T K N , which is a modification of an existing estimator), which are approximately unbiased when the error variances are known. However, they are biased when the error variance is unknown and needs to be estimated. This bias increases as ID increases. I show how this bias can be reduced. The results use semi-realistic simulations based on three real hyperspectral scenes. Despite this, when applied to the real scenes, R M T G and R M T K N are larger than expected. Possible reasons for this are discussed, including the presence of errors which are either deterministic, spectrally and/or spatially correlated, or signal-dependent. Possible future research into ID estimation in the presence of such errors is outlined. [ABSTRACT FROM AUTHOR]