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

We define the problem of maximal sensor integrity placement, that of locating sensors in n-dimensional grids with minimal vulnerability to enemy attack or sensor faults. We show a polynomial time algorithm for computing sensor integrity exists for sensors with unbounded ranges deployed over a 1D grid of points. We then present an integer linear programming (ILP) formulation for computing sensor integrity for unbounded range sensors over higher dimension grids.

As an attempt to solve some contemporary web information retrieval problems, a construction of a cooperative multiagent system is proposed. This. paper introduces the system and presents the use of a, unique aggregation and fusion technique that is employed to attain reliable delivery performance. The intelligent methodology to fuse agents' decisions, the team consensus approach, models the interaction and bring a society into a consensus. After each agent in the group gathers information relevant to a user's query, the group engages in an uncertainty estimation stage. This process allows each agent to assess its self-uncertainty and the conditional uncertainties of other agents. The procedure facilitates the computation of a weighting scheme that operates recursively on information collected by these agents until the group reaches a consensus. Whenever a new task is received, the uncertainty estimates of agent are updated and used to compute a new weighting scheme.

The joint probabilistic data association (JPDA) filter has a very good tracking performance in dense targets and heavy clutter environments. However, the JPDA filter also has a huge computer load and tends to combine neighboring tracks. In this paper, an improved JPDA algorithm is presented. The main feature of our method is improving the performance of the JPDA algorithm by improving the performance of the tracking gate. The effectiveness of this method is assessed by mathematical analysis.

In some information fusion processes, the incomplete information set can be naturally mapped into a belief theory information set and a Bayesian probability theory information set. For decision making, the mapping of the belief theory fusion results represented by the basic belief assignment to a probability set is accomplished via a pignistic probability transform. This article introduces the inverse pignistic probability transforms (IPPT) that map the posteriori probabilities into the belief function theories, basic belief assignments. Also introduced are two infinite classes and some finite classes of mapping the posteriori probability results to the basic belief assignment of the belief theory.

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).

Littoral tracking refers to the tracking of targets on land and in sea near the boundary of the two regions. A ground-moving target continues to move on land and can not enter the sea. Similarly, a sea-moving target moves in the sea and the land serves as an infeasible region. Enforcing infeasible regions or hard constraints in the framework of the Kalman filter or interacting multiple model (IMM) estimator is not natural. However, these hard constraints can be easily enforced using the particle filter algorithm. We formulate the littoral tracking problem as a joint tracking and classification problem, where we assign a target class for each isolated land or water region. We use a reflecting boundary condition to enforce the region constraint. We demonstrate this concept for a single target using the airborne ground moving target indicator measurements. Numerical results show that the proposed algorithm produces robust classification probabilities using kinematic measurements.

Spectral data provide opportunities to discriminate targets from background clutter and to detect partially concealed objects. However, data with high spatial resolution generally are not synoptic scale (hundreds of kilometers). By analyzing courser resolution synoptic imagery, high-resolution sensors can then be cued to areas of potential targets. Multispectral images (Landsat 7) are combined with ancillary data-lines of communication, digital elevation models (DEMS), etc.-in order to characterize the scene of interest. Two scenes were examined: regions of the Balkans (Kosovo, Bosnia-Herzegovina, Montenegro, and Serbia), and Iraq. Three data products result from this fusion of data sources: (1) land cover classification, (2) trafficability analysis, and (3) "hide area" delineation. In addition, the fusion of imagery with elevation models provides a beneficial perspective to the analyst. The classification is a thematic map of the different land cover types produced through a combination of supervised and unsupervised means. Trafficability may be dependent on a number of factors including land cover type, vegetation density, soil moisture content, and access to major roads or navigable rivers. In addition to land cover-based trafficability analysis, terrain-based trafficability uses DEM-derived slope information to determine vehicle accessibility. Determination of "hide areas" may be another important product. These can be defined by their proximity to the forest perimeter, access to roads, and the underlying terrain. As a result of these analyses carried out on the synoptic scale, the collective size of the areas of interest provided to the next sensor in the intelligence chain may be greatly reduced.

Ongoing efforts at our laboratory have targeted the development of techniques for fusing medical imagery of various modalities (i.e. MRI, CT, PET, SPECT, etc.) into single image products. Past results have demonstrated the potential for user performance improvements and workload reduction. While these are positive results, a need exists to address the three-dimensional nature of most medical image data sets. In particular, image fusion of three-dimensional imagery (e.g. MRI slices) must account for information content not only within a given slice but also across adjacent slices. In this paper, we describe extensions made to our 2D image fusion system that utilize 3D convolution kernels to determine locally relevant fusion parameters., Representative examples are presented for fusion of MRI and SPECT imagery. We also present these examples in the context of a GUI platform under development aimed at improving user-computer interaction for exploration and mining of medical data.

In this paper, a new Bayes fusion algorithm with the Parzen window method, which introduces the non-parameter estimation method of partition recognition into traditional Bayes fusion criterion, is propose. During the process of fusion, which is a repetitious and iterative process, conditional probability density is continuously modified and learned using the Parzen window method, and the global decision is obtained at the fusion center under the bayes decision criterion. In the practical application, the method has been successfully applied into the temperature fault detection and diagnosis system of hydroelectric simulation system of J. Fengman. The analysis of data indicates that the improved algorithm takes precedence over the traditional Bayes criterion.

The modern day challenges posed by terrorism and crime mean that we must make better use of the data we collect. We need to create a process that can transform different types of explicit and structured data into actionable knowledge. We need to be able to create unified infrastructure that will support a single query across all of the data sources. This unified infrastructure would allow us to integrate different collection stove pipes, including: text, structured data, images, faxes, audio, and video. To create this unified structure, we transform the collection stovepipes into sets of derived data that are integrated with structured data. Knowledge discovery tools are used over the entire set of collected data. This paper will present some of the components that we have been using to create knowledge infrastructures as well as the types of analysis required by our clients.