Polarimetric scattering models are developed to predict the detectability of surface-laid landmines. A specular polarimetric model works well only under the condition that there is either no sunlight or the sun is not close to the specular reflection direction. Moreover, this model does not give insight why certain man-made objects like landmines give a higher polarimetric signature than natural background. By introducing a polarimetric bidirectional reflectance distribution function (BRDF) the specular model is extended. This new model gives a better prediction of the polarimetric signature and gives a close match to the measurements of landmines with different casings as well as the sand background. The model parameters indicate that the landmines have a lower surface roughness and a higher refractive index, which is the reason why these objects are detectable from the background based on their polarimetric signature.
The potential for performance improvements through sensor fusion is explored for two electro-optical (EO) imaging sensors: a passive thermal IR camera and an active polarimetric system. Tests of decision-level fusion using a small data set (roughly 60 mine signatures) suggest that a significant performance improvement can be obtained by using an AND fusion approach. The source of this improvement derives from correlation among the sensors. Specifically, the sensors exhibit a strong positive correlation when a mine is present, and a negligible correlation when viewing clutter. The observed improvement is independent of the local ground clutter, but it depends strongly on the decision thresholds used for the individual sensors.
Algorithms are presented for detecting surface mines using multi-spectral data. The algorithms are demonstrated using visible and MWIR imagery collected at Fort A.P. Hill, VA under a variety of conditions. For imagery with a resolution of a few centimeters there is significant correlation in the clutter. Using a first-order Gauss Markov random field model for the clutter, an efficient pre-whitening filter is proposed. A significant improvement in detection is demonstrated as a result of this whitening. Further improvement in the detection of specific mine types is demonstrated by using a random signal model with a known covariance matrix. That approach leads to an estimator-correlator formulation, in which the random signature estimate is the output of a Wiener filter. It is suggested that by fusing the output of a bank of such filters one could improve detection of all mine types.
We investigate the potential for improving land mine detection by fusing data from ground penetrating radars (GPRs) and sensors of acoustically induced soil motion. We present a case study involving data from the SRI forward-looking radar and a laser Doppler vibrometer (LDV) system developed by the University of Mississippi. The LDV sensor detects acoustically induced soil vibrations, while the GPR detects scattering from dielectric discontinuities or metal objects in the soil. Since these sensors exploit different target physical properties, it is reasonable to expect a benefit in fusion. Although the sensors are relatively new, the LDV and GPR data exhibit evidence for complementarity, in that the GPR is significantly better at detecting metal mines, while the LDV is somewhat better at detecting plastic mines. Decision-level fusion is shown to improve performance. A simple OR fusion approach is found to perform similarly to an optimum hard decision fusion algorithm.
Many aspects of a buried mine's thermal IR signature can be predicted through physical models, and insight provided by such models can lead to better detection. Several techniques for exploiting this information are described. The first approach involves ML estimation of model parameters and followed by classification of those parameters. We show that this approach is related to an approximate evaluation of an integral over the parameters that arises in a Bayesian formulation. This technique is compared with a generalized likelihood ratio test (GLRT) and with computationally efficient, model-free approaches, in which soil temperature data are classified directly. The benefit of using the temporal information is also investigated. Algorithm performance is illustrated using broadband IR imagery of buried mines acquired over a 24 hour period. It is found that the detection performance at a suitably selected time is comparable to the performance achieved by processing all times. The performance of the GLRT, for which detection is based only on the residual error, is inferior to a classifier using the parameters.
KEYWORDS: Sensors, Sensor fusion, Sensor performance, Performance modeling, Land mines, Mining, Data acquisition, Systems modeling, Data fusion, Electromagnetic coupling
Simple theoretical models can be constructed to study the behavior of sensor-fused systems using idealized sensor suites. Models are available for feature-level and decision-level fusion, both of which are now being used with demining sensors. These models are attractive as design tools and for estimating the expected performance of new sensor suites, since their performance can be evaluated with relatively little effort. In this paper we review some simple idealized models and their predictions for fused system performance. The data produced by demining sensors are often correlated, and the effect of correlation is explored for both feature-level and decision-level fusion.
An image processing technique is described for the detection of miens in RI imagery. The proposed technique is based on a third-order neural network, which processes the output of a wavelet packet transform. The technique is inherently invariant to changes in signature position, rotation and scaling. The well-known memory limitations that arise with higher-order neural networks are addressed by (1) the data compression capabilities of wavelet packets, (2) protections of the image data into a space of similar triangles, and (3) quantization of that 'triangle space'. Using these techniques, image chips of size 28 by 28, which would require 0(109) neural net weights, are processed by a network having 0(102) weights. ROC curves are presented for mine detection in real and simulated imagery.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.