MENU

   Publications Home

   C4I Home
   Center Overview

   C4I People

   Objectives
   Programs
   Funding & Support
   Industry Partners

   Publications
   Recordings

   C4I Events
   News

   Internet Conference

C4I Publication Abstracts
Sensing and Fusion

C3I-4001

Polarimetric Subspace Target Detector
for SAR Data Based on the Huynen Dihedral Model

Authors: Vic Larson and Les Novak

Two new polarimetric subspace target detectors are developed based on a dihedral signal model for bright peaks within a spatially extended target signature. The first is a coherent dihedral target detector based on the exact Huynen model for a dihedral. The second is a noncoherent dihedral target detector based on the Huynen model with an extra unknown phase term. Expressions for these polarimetric subspace target detectors are developed for both additive Gaussian clutter and more general additive spherically invariant random vector (SIRV) clutter including the K-distribution. For the case of Gaussian clutter with unknown clutter parameters, constant false alarm rate (CFAR) implementations of these polarimetric subspace target detectors are developed. The performance of these dihedral detectors is demonstrated with real millimeter-wave fully polarimetric SAR data. The coherent dihedral detector which is developed with a more accurate description of a dihedral offers no performance advantage over the noncoherent dihedral detector which is computationally more attractive. The dihedral detectors do a better job of separating a set of tactical military targets from natural clutter compared to a detector that assumes no knowledge of the polarimetric structure of the target signal.

SPIE Conference, Orlando, FL, April 1995.

C3I-4002

Joint Spatial-Polarimetric Whitening Filter
to Improve SAR Target Detection Performance for
Spatially Distributed Targets

Authors: Vic Larson, Les Novak, and Clay Stewart

In this paper, we present a new spatial-polarimetric whitening filter (SPWF) detector. The SPWF detector is a straightforward extension of the polarimetric whitening filter (PWF) detector. The SPWF detector can be derived by applying the generalized likelihood ratio test to the detection problem when the clutter is additive complex Gaussian and the target signal is completely unknown. The SPWF detector is appropriate when the target energy is not localized in a single pixel (or polarimetric band) and the distribution of the target energy is completely unknown. We have found that the SPWF detector performs well against SAR data when it is applied with a small 2 x 2 pixel target window. In the SAR data, individual bright target scatterers are usually distributed over a few pixels, but have an unknown distribution. Often the bright scatterers of the target have high energy pixels surrounding the most dominant pixel. The SPWF detector performs better than the PWF and other single pixel detectors because it makes use of the energy in pixels surrounding the brightest pixel.

SPIE Conference on Algorithms for Synthetic Aperture Radar, Vol. 2230-32, pp. 285-301, Orlando, FL, April 6-7, 1994.

C3I-4003

Tracking And Fusion Using Multiple MTI Sensors

Author: Kuo-Chu Chang

Multi-sensor tracking and data fusion deals with combining data from various sources to arrive at an accurate assessment of a situation. Difficulties in performing multi-sensor tracking and fusion include not only ambiguous data, but also disparate data sources. The tracking and data association problem is further complicated by the facts that the target may not be detected by some sensors, dense false alarms and clutters may be present, and the target model may not be known exactly. In this paper, a multitarget tracking problem that involves data obtained from multiple MTI sensors is considered. A tracking and fusion algorithm that takes into account the uncertainties in both data origin and target dynamics under the clutter environment is presented.

IFAC Journal, Sept., 1994.

C3I-4004

Multiple Intelligence Correlation and Fusion
with Bayesian Networks

Author: Kuo-Chu Chang

Bayesian networks technology has good potential for applications in multi-source correlation and fusion. It introduces a set of representation techniques for encoding uncertain beliefs using probability theory and reasoning techniques for drawing inferences from such representations. The technology has been successfully applied both to tasks of assessment under uncertainty and tasks of decision-making under uncertainty. Multi-source intelligence fusion deals with combining data from various sources to arrive at an accurate assessment of the situation. It involves the integration, filtering, and correlation of information from diverse sources for the purpose of situation assessment, planning, making decisions, or improving system performance. Multi-source fusion in difficult situations requires the use of models to extract target information from the data. Bayesian networks show great promise for performing this function since they can be used to represent complicated probabilistic relationships between variables of interest. Furthermore, many efficient algorithms have been developed for drawing inferences from the evidence. In this paper, we consider the problem of multiple intelligence correlation and fusion using Bayesian networks. Particularly, the problem of fusing two intelligent data, ELINT and COMINT, for the purpose of target identification is presented. The issues of domain knowledge acquisition, network construction, and algorithm selection are also discussed.

Data Fusion Symposium, Oct., 1994.

C3I-4005

A Neural Network Approach for High Resolution
Target Classification

Authors: Kuo-Chu Chang and Yi-Chuan Lu

We proposed an improved version of the Self Organizing Feature Map/Learning Vector Quantitization (SOFM/LVQ) classifier currently used in an ATR system for SAR imagery. This classifier was originally designed to construct a small number of templates to represent a set of targets with different orientations. The classifier accepts an input of a target, computes distances of this data with those representative templates, and then classifies this data to the target class with the shortest distance. In this paper, we focus on the issue of how to identify and reject data from targets outside the given data set, such as man-made clutters. To reject clutters, we propose two discrimination functions, distance and entropy measures. With the distance discriminator, we obtained very good classification performance when all data are from the given target sets. However, the simple distance measure produces poor classification results when unknown targets such as natural or manmade clutters are present and when each target is represented by a small number of templates. We correct this deficiency by incorporating an entropy measure into the original classifier. With this entropy discriminator, our system rejects a majority of the false alarms while maintaining a high correct classification rate with relatively few templates for each target. Although this system was tested on real ISAR data and showed very good performance, the data was obtained from "turntable" experiments with a fixed depression angle and known target location. One future research direction is to test this algorithm with real "field" SAR data and study the robustness of the system.

SPIE, April, 1995.

C3I-4006

Multitarget Tracking with Adaptive Detection Thresholds

Author: Kuo-Chu Chang

The idea of adjusting the detection thresholds adaptively to enhance the performance of an overall tracking system has been an important research area studied in the tracking community for the last ten years. However, most of the previous work was developed for single target environments where a simple algorithm such as nearest neighbor (NN) or Probabilistic Data Association (PDA) filter was assumed for use in the tracking system. In this paper, we study the issues of adaptive detection thresholds based on the assumption that an optimal assignment algorithm is adopted for a multitarget and cluttered environment. This research is motivated by an important earlier work which makes the analytical evaluation of the optimal assignment algorithm possible. The performance measures considered for determining detection thresholds are the correct association probability and the expected estimation error. The analytical results obtained in this paper represent the upper bound of the tracking performance and can be used for designing and evaluating a tracking system.

ACC, Seattle, June, 1995 and IEEE Transactions on AES, June, 1995.

C3I-4007

A Greedy Assignment Algorithm and
Its Performance Evaluation

Authors: Kuo-Chu Chang and X. Zhao

It is well known that the performance of multiple-hypothesis approaches for multi-target tracking are near-optimal and have gained popularity since the pioneer work of Reid. In these approaches, all feasible data association hypotheses between measurements and targets are formed, evaluated, and maintained. Although they can handle complex target and sensor models and include track initiating and continuation in one framework, they require huge amounts of computing resources, both time and memory, especially under dense target and clutter environments.

It has been proven that a general form of a multiple-hypothesis tracking algorithm is in fact optimal under a given set of conditions if unlimited computing resources, both time and memory, are available. However, in practice, due to limited resources, suboptimal hypothesis management schemes such as pruning and combining are needed to make the algorithms feasible. Moreover, in order to determine which hypothesis to prune, all feasible hypotheses need to be generated and their associated probabilities evaluated. Unfortunately, as mentioned above, enumerating all the hypotheses is in general computationally unfeasible. It is therefore desirable to be able to determine the "N-best" hypotheses without first generating all of them. In fact, this has become the central issue of several recent research foci.

In the past, an approximate solution based on a heuristic approach was proposed, but it turns out to be very computationally inefficient. Furthermore, two exact algorithms were proposed which compute the N-best hypotheses. They start with the assumption that the best hypothesis is obtained first using a classical assignment algorithm such as the Hungarian algorithm. These algorithms then proceed to find the next N-1 best hypotheses without enumerating all the hypotheses. In fact, the run time of one of the algorithms proposed is proportional to N4 including the time to obtain the initial best hypothesis.

In this paper, we propose a suboptimal heuristic algorithm called "greedy branch and bound," where we determine approximately the best N hypotheses in a heuristic but efficient manner. The algorithm first orders the measurements based on the association likelihoods between the measurements and tracks. The "greedy" algorithm then expands the association hypotheses in a measurement-oriented manner based on the order obtained. The hypothesis expansion process is similar to the branching of a tree. The accumulated log-likelihoods of the current expanded branches are used to determine which branch to expand next in a "best-first" manner. The expansion process terminates when a specified number of hypotheses has reached the bottom of the tree and no other unexpanded branch has a higher accumulated "score." The algorithm is very simple and fast; however, no optimality is guaranteed. On the other hand, this trade-off may be favorable in a dense environment since "optimal" assignment may not be practical or necessary. The algorithm is tested extensively based on simulated data. Performance results show the effectiveness of the algorithm.

ACC, Seattle, June, 1995.

C3I-4008

Symbolic Probabilistic Inference with both Discrete
and Continuous Variables

Authors: Kuo-Chu Chang and Robert Fung

The importance of resolving general queries in Bayesian networks has been the focus of attention in recent research on the Symbolic Probabilistic Inference (SPI) algorithm. SPI applies the concept of dependency-directed backward search to probabilistic inference, and is incremental with respect to both queries and observations. Unlike traditional Bayesian network inferencing algorithms, the SPI algorithm is goal directed, performing only those calculations that are required to respond to queries. Research to date on SPI applies to Bayesian networks with only discrete-valued variables or only continuous variables (linear Gaussian), and does not address networks with both discrete and continuous variables.

In this paper, we extend the SPI algorithm to handle Bayesian networks made up of both discrete and continuous variables (SPI-DC). The only topological constraint of the networks is that the successors of any continuous variable must be continuous variables as well. In order to arrive at an exact analytical solution, the relationships between the continuous variables are restricted to be "linear Gaussian'". With new representation, SPI-DC modifies the three basic SPI operations: multiplication, summation, and substitution. However, SPI-DC retains the framework of the SPI algorithm, namely building the search tree and recursive query mechanism, and therefore retains the goal-directed and incrementality features of SPI.

To be published in IEEE Transaction on Systems, Man, and Cybernetics, June, 1995.

C3I-4009

Compact Acousto-optic Processor
for Synthetic Aperture Radar Image Formation

Author: Michael W. Haney, ECE Department, George Mason University

A compact implementation of a real-time acousto-optic synthetic aperture radar (SAR) imager is described. The architecture generates SAR imagery by decomposing the required 2-D integration into a cascade of two 1-D integrations, in a manner similar to that used in computed tomography for medical imaging applications. To achieve low power consumption, the required integrations are performed in the analog optical domain, with a crossed Bragg cell configuration, using a combination of spatial and temporal integration of optical signals. The time integrating and space integrating modules of the system are coupled via a common path interferometer. Subimages are formed on a 2-D CCD detector array that is rotated during image formation to avoid a computationally difficult digital interpolation operation. The complex-valued subimages are combined in a digital frame buffer for dynamic range enhancement before being displayed. To accommodate needed flexibility, the required filter functions are calculated in a digital controller and downloaded through the Bragg cells. Results of a laboratory demonstration are presented. Performance projections suggest that the architecture may offer an advantage over an all- electronic approach for high resolution applications which are severely constrained in power consumption.

Invited paper in Proceedings of IEEE, Vol. 82, No. 11, November 1994.

C3I-4010

Compact Time- and Space-Integrating SAR Processor:
Design and Development Status

Authors: Michael W. Haney, ECE Department, George Mason University; James J. Levy, Marc P. Christensen, and Robert R. Michael, BDM Federal, Inc.; Michael W. Mock, Loral Defense Systems

Progress toward a flight demonstration of the acousto-optic time- and space-integrating real-time SAR image formation processor program is reported. The concept overcomes the size and power consumption limitations of electronic approaches by using compact, rugged, and low-power analog optical signal processing techniques for the most computationally taxing portions of the SAR imaging problem. Flexibility and performance are maintained by the use of digital electronics for the critical low- complexity filter generation and output image processing functions. The results reported include tests of a laboratory version of the concept, a description of the compact optical design that will be implemented, and an overview of the electronic interface and controller modules of the flight-test system.

Proceedings of SPIE, Vol. 2236, April, 1994

C3I-4011

Evolving Neural Networks for
Video Attitude and Height Sensor

Authors: Zhixiong Zhang and Kenneth J. Hintz

The development of an on-board Video Attitude and Height Sensor (VAHS) which is used to measure the height, roll, and pitch of an airborne vehicle at low altitude, is presented. The VAHS consists of a downlooking TV camera and two orthogonal sets of laser diodes (total four laser diodes) producing a structured light pattern. Although the height, roll, and pitch can be determined by measuring the locations of the dots in the image, in practice it is very difficult to precisely align the laser diodes and the TV camera. Moreover it is also very hard to obtain accurate camera parameters, because of its various nonlinear distortions. An approach which uses layered neural networks (NNs) to map the locations of the dots in the image to the height, roll, and pitch of the airborne vehicle, is presented here. Amorphous NNs have also been evolved by genetic algorithms (GAs) with mixed results. Some simulation results of these experiments are presented.

Presented at SPIE's International Symposium on Aerospace/Defense Sensing & Control and Dual-Use Photonics '95, Orlando FL, April 1995.