Evaluating backgrounds for subpixel target detection: when closer isn't better

2015

SPIE Proceedings

We can improve the detection of targets and anomalies in a cluttered background by more effectively estimating that background. With a good estimate of what the target-free radiance or reflectance ought to be at a pixel, we have a point of comparison with what the measured value of that pixel actually happens to be. It is common to make this estimate using the mean of pixels in an annulus around the pixel of interest. But there is more information in the annulus than this mean value, and one can derive more general estimators than just the mean. The derivation pursued here is based on multivariate regression of the central pixel against the pixels in the surrounding annulus. This can be done on a band-by-band basis, or with multiple bands simultaneously.For overhead remote sensing imagery with square pixels, there is a natural eight-fold symmetry in the surrounding annulus, corresponding to reflection and right angle rotation. We can use this symmetry to impose constraints on the estimator function, and we can use these constraints to reduce the number or regressor variables in the problem. This paper investigates the utility of regression generally -and a variety of different symmetric regression schemes particularly -for hyperspectral background estimation in the context of generic target detection.

Section: Discussionmentioning

confidence: 81%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Symmetrized regression for hyperspectral background estimation

2015

SPIE Proceedings

“…[30][31][32][33][34] The idea of local means was extended to include more general local estimators based on functions of the pixels in the annulus. [18][19][20]35 background pixels in image subpixel target in image …”

Section: Residual Spacementioning

confidence: 99%

Graph-based and statistical approaches for detecting spectrally variable target materials

Ziemann

2016

SPIE Proceedings

In discriminating target materials from background clutter in hyperspectral imagery, one must contend with variability in both. Most algorithms focus on the clutter variability, but for some materials there is considerable variability in the spectral signatures of the target. This is especially the case for solid target materials, whose signatures depend on morphological properties (particle size, packing density, etc.) that are rarely known a priori. In this paper, we investigate detection algorithms that explicitly take into account the diversity of signatures for a given target. In particular, we investigate variable target detectors when applied to new representations of the hyperspectral data: a manifold learning based approach, and a residual based approach. The graph theory and manifold learning based approach incorporates multiple spectral signatures of the target material of interest; this is built upon previous work that used a single target spectrum. In this approach, we first build an adaptive nearest neighbors (ANN) graph on the data and target spectra, and use a biased locally linear embedding (LLE) transformation to perform nonlinear dimensionality reduction. This biased transformation results in a lower-dimensional representation of the data that better separates the targets from the background. The residual approach uses an annulus based computation to represent each pixel after an estimate of the local background is removed, which suppresses local backgrounds and emphasizes the target-containing pixels. We will show detection results in the original spectral space, the dimensionality-reduced space, and the residual space, all using subspace detectors: ranked spectral angle mapper (rSAM), subspace adaptive matched filter (ssAMF), and subspace adaptive cosine/coherence estimator (ssACE). Results of this exploratory study will be shown on a ground-truthed hyperspectral image with variable target spectra and both full and mixed pixel targets.

“…One final item on this "to-do" list would be to employ local, instead of global, mean subtraction. Here, the mean m at a given pixel x is estimated from a local annulus surrounding that pixel; this idea been employed already in a variety of studies, [44][45][46][47][48][49][50][51][52][53][54][55] and Appendix B shows how the likelihood function is altered for generic target detection. One of the earliest advocates of local mean subtraction was the original RX anomaly detection paper by Reed and Yu, 44 which also remarked that the distribution of the residual x − m is often more nearly Gaussian than the unsubtracted x values.…”

Section: Model Distributionmentioning

confidence: 99%

Anisotropic background models for spectral target detection

Algorithms, Technologies, and Applications for Multispectral and Hyperspectral Imaging XXIX

2023

Algorithms are derived for detecting targets in cluttered backgrounds, where the background is modeled as a product of univariate distributions independently fit to each of the principal component projections. Thus, fatter-than-Gaussian tails are fit to the data, with a different fatness parameter for each principal component. Comparisons are made to elliptically-contoured distributions (which, unlike these product distributions, are isotropic in the whitened space), including the multivariate t and the Gaussian. Numerical experiments are performed on hyperspectral data from the SHARE 2012 exercise, with target detection performance evaluated on both actual and simulated targets. Both direct and residual data are considered, with the residual data obtained from local background subtraction -these residual data are found to exhibit not only lower variance, but qualitatively different tail statistics. More direct target-agnostic measures are also employed to asses how well these models fit the different kinds of background clutter.