Proceedings of the Fifth International Conference on Information Fusion. FUSION 2002. (IEEE Cat.No.02EX5997)

In support of the AFOSR program in Information Fusion, the CNS Technology Laboratory at Boston University is developing and applying neural models of image and signal processing, pattern learning and recognition, associative learning dynamics, and 3D visualization, to the domain of Information Fusion for Image Analysis in a geospatial context. Our research is focused by a challenge problem involving the emergence of a crisis in an urban environment, brought on by a terrorist attack or other man-made or natural disaster. We aim to develop methods aiding preparation and monitoring of the battlespace, deriving context from multiple sources of imagery (high-resolution visible and low-resolution hyperspectral) and signals (GMTI from moving vehicles, and ELINT from emitters). This context will serve as a foundation, in conjunction with existing knowledge nets, for exploring neural methods in higher level information fusion supporting situation assessment and creation of a common operating picture (COP).

In this paper tracking multiple spawning targets with multiple finite-resolution sensors is presented with emphasis on the measurement-to-track association with possibly unresolved measurements. The goal is to initialize new tracks of spawned targets before they are resolved from the mother platform's so that one has the ability to carry out early discrimination when they become resolved. The multiple scan data association problem is first formulated as a multidimensional assignment problem with explicit new constraints for the unresolved measurements. The top M hypotheses tracking (TMHT) is presented where the state estimates and their covariances are modified based on the M best hypotheses through the assignment solutions. A modification to the assignment problem is developed that leads to a linear programming (LP) where the optimal solution can be a noninteger in [0, 1]. The fractional optimal solution is interpreted as (pseudo)probabilities over the N-1 frame sliding window. The best hard (binary) decision assignment solution and the M best via TMHT are compared with the soft decision (LP) solution for two-dimensional tracking scenarios with various sensor configurations. Based on the simulation results, the soft assignment approach has better track maintenance capability than the single best hard assignment and a performance close to TMHT. Its computational load is slightly higher than the single best hard assignment an much lighter than TMHT.

Nowadays, there is a huge volume of information on the Web, which is disseminated to users in a chaotic way. In order to be easily accessed, the information must be clustered and classified in appropriate knowledge areas. Thus, many heavily visited sites or portals attempt to unify the access to multiple information sources, providing by this way classification of information. The paper proposes a system, aiming to classify e-commerce sites according their Web content. This system can be implemented for automatic knowledge segmentation in a portal or in a search engine repository. The system performance reached 96% in the first test sets, after the learning phase. However, the performance significantly increases (up to 98%) as the number of test sets increases.

Information fusion based on Kalman filtering often suffers from the lack of knowledge about cross correlations between the noise-corrupted signal sources. Covariance intersection filtering provides a general framework for information fusion with incomplete knowledge about the signal sources since it yields consistent estimates for any degree of cross correlation. However, covariance intersection filtering requires optimization of a nonlinear cost function which is a significant drawback with respect to computational complexity. Therefore, a fast covariance intersection algorithm is developed and investigated based on simulation results.

This paper presents the design, development, and validation of a nonlinear least square estimation scheme applied to star tracker noise extraction and identification. The paper is the by-product of a Post-Launch Test (PLT) tool development effort conducted by two independent teams, Swales/NASA and Boeing. The main objective is to have a set of tools ready to provide on-orbit support to the GOES N-Q Program. GOES N-Q employs a stellar inertial attitude determination (SIAD) system that achieves high precision attitude estimation by processing attitude and rate data provided by multiple star trackers (ST) and an inertial reference unit (IRU), respectively. The key component of SIAD is the ST. The ST's star position vector is corrupted by three major noise sources: temporal noise (TN), high spatial frequency noise (HSF), and low spatial frequency (LSF) noise. The last two noise sources are not while and correlated. As a result, the performance of the SIAD filter is no longer optimal, causing the reconstructed attitude knowledge to potentially satisfy requirements with a narrow margin. This tight margin is critical and may affect the GOES N-Q mission, particularly the Image Navigation and Registration (INR) system performance. The PLT toolset is expected to provide the capability to mitigate this potential problem during PLT time.

In tracking applications, target dynamics is usually modeled in the Cartesian coordinates, while target measurements are directly available in the original sensor coordinates. Measurement conversion is widely used such that the Kalman filter in the Cartesian coordinates can be applied. A number of improved measurement-conversion techniques have been proposed recently. However, they have fundamental limitations, resulting in performance degradation, as pointed out in Li and Jilkov (2001) of a recent survey. This paper proposes a recursive filter that is theoretically optimal in the sense of minimizing the mean-square error among all linear unbiased filters in the Cartesian coordinates. The proposed filter is free of the fundamental limitations of the measurement-conversion approach. Results of an approximate implementation are compared with those obtained by two state-of-the-art conversion techniques. Simulation results are provided.

We propose a multiresolution fusion algorithm which combines aspects of region and pixel-based fusion. We use multiresolution decompositions to represent the input images at different scales, and introduce a multiresolution/multimodal segmentation to partition the image domain at these scales. The basic idea is to use this segmentation to guide the fusion process. A region-based multiresolution approach allows us to consider low-level as well as intermediate-level structures, and to impose data-dependent consistency constraints based on spatial, inter and intra-scale dependencies.

Decision making is always a difficult assignment for everyone in every context and more information eases the process and makes it more accurate. More data does not always mean more information. Rough sets alone can be adopted to handle hierarchical structure in the family of criteria using decision rules based on preference model. The rules are induced from the preferential information given by the Decision Maker (DM) in the form of examples of decisions. The existing rough set techniques are applicable in information systems where condition and decision attributes are distinguishable. But, in general, the information are available in the form of a data table, called an information system or knowledge representation system, where rows are labeled by alternatives and columns by attributes. In such cases, the condition and decision attributes are not distinguishable. The present paper offers a solution for making decisions in an efficient way when enough data are available, relevant or irrelevant. In a Multiple Criteria Decision Analysis (MCDA) problem, determination of weights is an important aspect of the Analytic Hierarchy Process (AHP). Based on rough set concepts, a model, viz., the interval AHP model with interval data, for determination of the weights has been considered here. According to the necessary and possibility concepts of rough set, we obtain the lower and upper evaluation models, respectively. After obtaining the local interval weights, a method is proposed for calculating the global weights, and also for ranking the alternatives. The proposed method will be applicable for an information system as well as for a decision table.

The design of a generic mechanism for platform-based data fusion and sensor management is described. It is based on the duality between two types of agents, task agents and sensor agents. Task agents buy information from sensor agents that sell it, and sensors produce information that tasks consume. These buy and sell interactions occur very frequently for systems involving sensors that work with small action durations. When several sensors are available, the sell/buy interaction involve selection and scheduling of sensors. The task concept is similar to the old decision or OODA (observe/orient/decide/act) loop that has since long been used for understanding human participation in complex command and control problems. A task might be described as a tiny OODA loop, with predefined purpose and processing capability. There are numerous task types, each dedicated to a certain skill or sensor process. The design is evaluated in single and multiplatform applications.

The future success of the National Imagery and Mapping Agency (NIMA) and the National System for Geospatial Intelligence (NSGI) will hinge on the agency's ability to deliver timely, tailored, precise data, information and knowledge to its customers. NIMA must understand what information truly matters and foster an analytical environment that encourages creativity and excellence. Achieving this vision will require transformation in a wide spectrum of activities, including leveraging revolutionary information and knowledge fusion methods, techniques and technologies. Since September 11, 2001, NIMA has restructured operations to ensure information superiority and is continuing the transformation from being a producer of imagery and geospatial information, to being a provider of geospatial intelligence that will enable decision dominance for its customers. The goals delineated by the NIMA Strategic Intent are based on delivering predictive and actionable decision knowledge extracted from fused data and information from multiple intelligence sources into a complete geospatially-referenced common operational picture (COP). The requirements of information superiority and decision dominance present tremendous challenges for NSGI and NIMA, as well as to the entire intelligence community. To rise to these challenges, NIMA envisions leveraging the best commercial and government business, operational and acquisition practices, migration strategies, commercial and mission specific technologies.