Intelligent foraging, gathering and matching (I-FGM) has been shown to be an effective tool for intelligence analysts who have to deal with large and dynamic search spaces. I-FGM introduced a unique resource allocation strategy based on a partial information processing paradigm which, along with a modular system architecture, makes it a truly novel and comprehensive solution to information retrieval in such search spaces. This paper provides further validation of its performance by studying its behavior while working with highly dynamic databases. Results from earlier experiments were analyzed and important changes have been made in the system parameters to deal with dynamism in the search space. These changes also help in our goal of providing relevant search results quickly and with minimum wastage of computational resources. Experiments have been conducted on I-FGM in a realistic and dynamic simulation environment, and its results are compared with two other control systems. I-FGM clearly outperforms the control systems.

In future battle spaces, multiple disparate sensors and unmanned vehicles will be in simultaneous use and form ad hoc networks whose services collectively reason on the situation. These networks may come under attack by malignant devices sending false information. The network services must evaluate incoming information to determine if the information is relevant and trustworthy. Information Forensics services can accomplish this evaluation by interrogating the source. The competency of an interrogator can be quantified by the level of their questions. This paper will discuss the different levels of abstraction in learning and how they relate to networks that support active querying.

In this paper, we present the PEGASUS system. PEGASUS is an integrated news video search system with three major components: (1) user interface, where users can formulate search queries and browse the returned results; (2) server, which takes the queries from user interface, performs the searches and ranks the search results before returns them to the users; (3) data storage, which is composed of feature indexing system and video database. The PEGASUS system has the capability to allow users to perform fast multi-modality video search using video features, including features from both audio and visual portions of the videos. To search a target topic, the user first submits an initial query using the prior knowledge on the topic. Then, through a series of relevance feedback processes, a set of relevant video shots are returned by the system to the user. The user is able to further view the results using the video-on-demand (VoD) functionality of the system. The system has been constructed using over 45,000 news video shots, and it is available online of public access.

Fuzzy cognitive maps are an emerging technique for knowledge elicitation and data synthesis. The technique can capture the cause and effect relationships that subject matter experts believe to exist about a problem. A chief advantage of this method is that a common metric for different attributes does not need to be determined because states of attributes are compared to states of attributes. This is also a disadvantage because the map can only infer a qualitative state for a node of interest, not a numerical value. To overcome this limitation, nodes are modeled using fuzzy sets that are then propagated through the map. Borrowing techniques used in fuzzy control systems, the scaled fuzzy sets can then be used to yield a crisp numerical value for the attribute represented by the node.

Fully exploiting the intelligence community's exponentially growing data resources will require computational approaches differing radically from those currently available. Intelligence data is massive, distributed, and heterogeneous. Conventional approaches requiring highly structured and centralized data will not meet this challenge. We report on a new approach, Agent-Based Reasoning (ABR). In NIST evaluations, the use of ABR software tripled analysts' solution speed, doubled accuracy, and halved perceived difficulty. ABR makes use of populations of fine-grained, locally interacting agents that collectively reason about intelligence scenarios in a self-organizing, "bottom-up" process akin to those found in biological and other complex systems. Reproduction rules allow agents to make inferences from multi-INT data, while movement rules organize information and optimize reasoning. Complementary deterministic and stochastic agent behaviors enhance reasoning power and flexibility. Agent interaction via small-world networks - such as are found in nervous systems, social networks, and power distribution grids - dramatically increases the rate of discovering intelligence fragments that usefully connect to yield new inferences. Small-world networks also support the distributed processing necessary to address intelligence community data challenges. In addition, we have found that ABR pre-processing can boost the performance of commercial text clustering software. Finally, we have demonstrated interoperability with Knowledge Engineering systems and seen that reasoning across diverse data sources can be a rich source of inferences.

A wireless ad hoc sensor network (WSN) is a configuration for area surveillance that affords rapid, flexible deployment in arbitrary threat environments. There is no infrastructure support and sensor nodes communicate with each other only when they are in transmission range. To a greater degree than the terminals found in mobile ad hoc networks (MANETs) for communications, sensor nodes are resource-constrained, with limited computational processing, bandwidth, memory, and power, and are typically unattended once in operation. Consequently, the level of information exchange among nodes, to support any complex adaptive algorithms to establish network connectivity and optimize throughput, not only deplete those limited resources and creates high overhead in narrowband communications, but also increase network vulnerability to eavesdropping by malicious nodes. Cooperation among nodes, critical to the mission of sensor networks, can thus be disrupted by the inappropriate choice of the method for self-organization. Recent published contributions to the self-configuration of ad hoc sensor networks, e.g., self-organizing mapping and swarm intelligence techniques, have been based on the adaptive control of the cross-layer interactions found in MANET protocols to achieve one or more performance objectives: connectivity, intrusion resistance, power control, throughput, and delay. However, few studies have examined the performance of these algorithms when implemented with the limited resources of WSNs. In this paper, self-organization algorithms for the initiation, operation and maintenance of a network topology from a collection of wireless sensor nodes are proposed that improve the performance metrics significant to WSNs. The intelligent algorithm approach emphasizes low computational complexity, energy efficiency and robust adaptation to change, allowing distributed implementation with the actual limited resources of the cooperative nodes of the network. Extensions of the algorithms from flat topologies to two-tier hierarchies of sensor nodes are presented. Results from a few simulations of the proposed algorithms are compared to the published results of other approaches to sensor network self-organization in common scenarios. The estimated network lifetime and extent under static resource allocations are computed.

A new approach to neural networks is proposed, based on wireless interconnects (synapses) and cellular neurons, both software and hardware; with the capacity of 10¹⁰ neurons, almost fully connected. The core of the system is Spatio-Temporal-Variant (STV) kernel and cellular axon with synaptic plasticity variable in time and space. The novel large neural network hardware is based on two established wireless technologies: RF-cellular and IR-wireless.

Sensor network technology has enabled new surveillance systems where sensor nodes equipped with processing and communication capabilities can collaboratively detect, classify and track targets of interest over a large surveillance area. In this paper we study distributed fusion of multimodal sensor data for extracting target information from a large scale sensor network. Optimal tracking, classification, and reporting of threat events require joint consideration of multiple sensor modalities. Multiple sensor modalities improve tracking by reducing the uncertainty in the track estimates as well as resolving track-sensor data association problems. Our approach to solving the fusion problem with large number of multimodal sensors is construction of likelihood maps. The likelihood maps provide a summary data for the solution of the detection, tracking and classification problem. The likelihood map presents the sensory information in an easy format for the decision makers to interpret and is suitable with fusion of spatial prior information such as maps, imaging data from stand-off imaging sensors. We follow a statistical approach to combine sensor data at different levels of uncertainty and resolution. The likelihood map transforms each sensor data stream to a spatio-temporal likelihood map ideally suitable for fusion with imaging sensor outputs and prior geographic information about the scene. We also discuss distributed computation of the likelihood map using a gossip based algorithm and present simulation results.

The purpose of this research was to design and implement the functional requirements of three important distributed services in a secure mobile ad-hoc network. The three distributed services are described: lookup services, adaptation services and composition services. Further, research was required to implement security at various layers to enhance the overall security of the SAFEMITS network. This required an extensive analysis of the security features of lookup server which functions as the controller of the mobile ad hoc network. Finally a technique was designed to select a super node, and a performance test was performed using both the Windows and Linux operating systems.

Color is an important feature for object recognition in security and military applications. Unfortunately, color is sensitive to the environmental operating conditions so its use for automatic target recognition is often limited. Recently a number of research efforts have focused on techniques for developing algorithms to improve color constancy across images. Many of these approaches attempt to improve the color constancy of a particular type of surface area such as skin. In contrast, we present an approach that attempts to address color constancy of many surfaces across a wide range of external environmental conditions in the absence of direct knowledge of illumination. Our approach builds on existing techniques by using evolutionary learning to synthesize features that characterize the illuminations that influence perception of color. Once the illumination of each image in a collection is estimated, it can be used to map the colors in an image to the illumination conditions in any other image. This would allows us to take an image from that collection, transform its colors to reference colors that can then be combined with other types of features (e.g. geometrical, statistical, and textural) to cerate automatic target recognition systems that are relatively insensitive to their operating conditions. To demonstrate our technique, we process images of a parking area under a wide variety of seasonal weather conditions collected across large timescales of hours, days, and months.

In this paper, we have primarily discussed technical challenges and navigational skill requirements of mobile robots for traversability path planning in natural terrain environments similar to Mars surface terrains. We have described different methods for detection of salient terrain features based on imaging texture analysis techniques. We have also presented three competing techniques for terrain traversability assessment of mobile robots navigating in unstructured natural terrain environments. These three techniques include: a rule-based terrain classifier, a neural network-based terrain classifier, and a fuzzy-logic terrain classifier. Each proposed terrain classifier divides a region of natural terrain into finite sub-terrain regions and classifies terrain condition exclusively within each sub-terrain region based on terrain visual clues. The Kalman Filtering technique is applied for aggregative fusion of sub-terrain assessment results. The last two terrain classifiers are shown to have remarkable capability for terrain traversability assessment of natural terrains. We have conducted a comparative performance evaluation of all three terrain classifiers and presented the results in this paper.

The commonality of cortical architecture in perceptual, motor and higher cognitive areas of the brain suggests that one or just a few basic computational mechanisms underlie a variety of seemingly unrelated abilities. The map-seeking circuit (MSC) is a computational mechanism with plausible neuronal implementations which efficiently solves inverse transformation-discovery problems of the dimensionality found in vision, inverse kinematics, route-planning and other "brain-solvable" natural tasks. As in the brain, the cooperative interaction of MSCs operating in different domains allows efficient and robust solution to problem such as recognition of articulated objects. The algorithmic versions of MSC benefit from the same combinatorial efficiencies as the neuronal versions, making it a practical method for target recognition and other defense and security tasks. Several areas of application MSC are demonstrated.

Electric power distribution systems can be found almost everywhere, from ship power systems to data centers. In many critical applications, there is needed to maintain minimal operating capability under fault conditions. To carry out this goal it is necessary to develop energy distribution control techniques, which let implement a self-reconfigurable energy distribution system. This research project is looking at the implementation of multi-agent systems to develop a self-reconfigurable electric power distribution system. A prototype of a Multi-Agent system is proposed to reconfigure the system in order to maximize the number of served loads with highest priority. The shipboard power system is simulated in Matlab^TM Simulink^TM and the Multi-Agent System is implemented using Java programming language and JADE platform.

One of the key challenges facing the global war on terrorism (GWOT) and urban operations is the increased need for rapid and diverse information from distributed sources. For users to get adequate information on target types and movements, they would need reliable data. In order to facilitate reliable computational intelligence, we seek to explore the communication modulation tradeoffs affecting information distribution and accumulation. In this analysis, we explore the modulation techniques of Orthogonal Frequency Division Multiplexing (OFDM), Direct Sequence Spread Spectrum (DSSS), and statistical time-division multiple access (TDMA) as a function of the bit error rate and jitter that affect targeting performance. In the analysis, we simulate a Link 16 with a simple bandpass frequency shift keying (PSK) technique using different Signal-to-Noise ratios. The communications transfer delay and accuracy tradeoffs are assessed as to the effects incurred in targeting performance.

A significant challenge in robotics is providing a robot with the ability to sense its environment and then autonomously move while accommodating obstacles. The DARPA Grand Challenge, one of the most visible examples, set the goal of driving a vehicle autonomously for over a hundred miles avoiding obstacles along a predetermined path. Map-Seeking Circuits have shown their biomimetic capability in both vision and inverse kinematics and here we demonstrate their potential usefulness for intelligent exploration of unknown terrain using a multi-articulated linear robot. A robot that could handle any degree of terrain complexity would be useful for exploring inaccessible crowded spaces such as rubble piles in emergency situations, patrolling/intelligence gathering in tough terrain, tunnel exploration, and possibly even planetary exploration. Here we simulate autonomous exploratory navigation by an interaction of terrain discovery using the multi-articulated linear robot to build a local terrain map and exploitation of that growing terrain map to solve the propulsion problem of the robot.

Dynamic resource allocation for sensor management is a problem that demands solutions beyond traditional approaches to optimization. Market-based optimization applies solutions from economic theory, particularly game theory, to the resource allocation problem by creating an artificial market for sensor information and computational resources. Intelligent agents are the buyers and sellers in this market, and they represent all the elements of the sensor network, from sensors to sensor platforms to computational resources. These agents interact based on a negotiation mechanism that determines their bidding strategies. This negotiation mechanism and the agents' bidding strategies are based on game theory, and they are designed so that the aggregate result of the multi-agent negotiation process is a market in competitive equilibrium, which guarantees an optimal allocation of resources throughout the sensor network. This paper makes two contributions to the field of market-based optimization: First, we develop a market protocol to handle heterogeneous goods in a dynamic setting. Second, we develop arbitrage agents to improve the efficiency in the market in light of its dynamic nature.

The signal to interference plus noise ratio (SINR) performance of SAR imagery knowledge-aided (KA) airborne moving target indicator (AMTI) radar subjected to severely taxing environmental disturbances is investigated for radar-blind and radar-seeing highly compressed SAR imagery. Radar-seeing schemes are found to greatly outperform radar-blind techniques.

Intelligent medical systems based on supervised and unsupervised artificial neural networks are applied to the automatic visualization and classification of suspicious lesions in breast MRI. These systems represent an important component of future sophisticated computer-aided diagnosis systems and enable the extraction of spatial and temporal features of dynamic MRI data stemming from patients with confirmed lesion diagnosis. By taking into account the heterogenity of the cancerous tissue, these techniques reveal the malignant, benign and normal kinetic signals and and provide a regional subclassification of pathological breast tissue. Intelligent medical systems are expected to have substantial implications in healthcare politics by contributing to the diagnosis of indeterminate breast lesions by non-invasive imaging.

When an image of a 3-D scene is captured, only scene parts at the focus plane appear sharp. Scene parts in front of or behind the focus plane appear blurred. In order to create an image where all scene parts appear sharp, it is necessary to capture images of the scene at different focus levels and fuse the images. In this paper, first registration of multifocus images is discussed and then an algorithm to fuse the registered images is described. The algorithm divides the image domain into uniform blocks and for each block identifies the image with the highest contrast. The images selected in this manner are then locally blended to create an image that has overall maximum contrast. Examples demonstrating registration and fusion of multifocus images are given and discussed.

We propose a content-based 3D mosaic (CB3M) representation for long video sequences of 3D and dynamic scenes captured by a camera on a mobile platform. The motion of the camera has a dominant direction of motion (as on an airplane or ground vehicle), but 6 DOF motion is allowed. In the first step, a set of parallel-perspective (pushbroom) mosaics with varying viewing directions is generated to capture both the 3D and dynamic aspects of the scene under the camera coverage. In the second step, a segmentation-based stereo matching algorithm is applied to extract parametric representations of the color, structure and motion of the dynamic and/or 3D objects in urban scenes where a lot of planar surfaces exist. Multiple pairs of stereo mosaics are used for facilitating reliable stereo matching, occlusion handling, accurate 3D reconstruction and robust moving target detection. We use the fact that all the static objects obey the epipolar geometry of pushbroom stereo, whereas an independent moving object either violates the epipolar geometry if the motion is not in the direction of sensor motion or exhibits unusual 3D structures. The CB3M is a highly compressed visual representation for a very long video sequence of a dynamic 3D scene. More importantly, the CB3M representation has object contents of both 3D and motion. Experimental results are given for the CB3M construction for both simulated and real video sequences to show the accuracy and effectiveness of the representation.

An analytical differential equation model from a single simulation input and output data vector is derived. The derived model is analytically varied (real versus imaginary) to determine Critical, Sensitive, and Key parameters without the use of Design of Experiments (DOE).

Image exploitation algorithms for Intelligence, Surveillance and Reconnaissance (ISR) and weapon systems are extremely sensitive to differences between the operating conditions (OCs) under which they are trained and the extended operating conditions (EOCs) in which the fielded algorithms are tested. As an example, terrain type is an important OC for the problem of tracking hostile vehicles from an airborne camera. A system designed to track cars driving on highways and on major city streets would probably not do well in the EOC of parking lots because of the very different dynamics. In this paper, we present a system we call ALPS for Adaptive Learning in Particle Systems. ALPS takes as input a sequence of video images and produces labeled tracks. The system detects moving targets and tracks those targets across multiple frames using a multiple hypothesis tracker (MHT) tightly coupled with a particle filter. This tracker exploits the strengths of traditional MHT based tracking algorithms by directly incorporating tree-based hypothesis considerations into the particle filter update and resampling steps. We demonstrate results in a parking lot domain tracking objects through occlusions and object interactions.

Vast quantities of EO and IR data are collected on airborne platforms (manned and unmanned) and terrestrial platforms (including fixed installations, e.g., at street intersections), and can be exploited to aid in the global war on terrorism. However, intelligent preprocessing is required to enable operator efficiency and to provide commanders with actionable target information. To this end, we have developed an image plane tracker which automatically detects and tracks multiple targets in image sequences using both motion and feature information. The effects of platform and camera motion are compensated via image registration, and a novel change detection algorithm is applied for accurate moving target detection. The contiguous pixel blob on each moving target is segmented for use in target feature extraction and model learning. Feature-based target location measurements are used for tracking through move-stop-move maneuvers, close target spacing, and occlusion. Effective clutter suppression is achieved using joint probabilistic data association (JPDA), and confirmed target tracks are indicated for further processing or operator review. In this paper we describe the algorithms implemented in the image plane tracker and present performance results obtained with video clips from the DARPA VIVID program data collection and from a miniature unmanned aerial vehicle (UAV) flight.

Tracking an entity for a long duration allows the gathering of intelligence on a target. While the system comprises a collection of different elements (e.g., tracking, sensor tasking, etc.), the ability to track objects continuously over long periods rests on feature measurements that are collected "on-the-fly" and used to uniquely characterize the target of interest. These features are then used to track the target over extended periods of time and through situations in which the targets can be confused with other moving objects. The collecting of features helps support tracking the target when it becomes kinematically ambiguous with other objects. If the system is unable to avoid ambiguities between the target of interest and other moving objects, features collected post-ambiguity can be used to resolve the ambiguities. A collection of algorithms that model and attempt to resolve any association ambiguity between a target of interest and the tracks in the fusion and tracking database is required to accomplish this task. This module is referred to as the Tracked Object Manager (TOM) and forms the backbone of a system for the continuous tracking of high-value targets. The TOM utilizes the collected features to help correct track switches and, if appropriate, stitch tracks together to maintain continuous track on high-value targets. The algorithms are being incorporated into and evaluated using Toyon's Intelligence, Surveillance and Reconnaissance (ISR) simulation environment named SLAMEM.

In many cases, tracking ground targets can be formulated as a nonlinear filtering problem when terrain and road constraints are incorporated into system modeling and polar coordinate is used. Furthermore, when tracking ground maneuvering targets with an interacting multiple model (IMM) approach, a non-Gaussian problem exists due to an inherent mixing operation. A multirate interacting multiple model particle filter (MRIMM-PF) is presented in this paper to effectively solve the problem of nonlinear and non-Gaussian tracking, with an emphasis on computational savings.

This paper presents the mathematical theory and procedure for comparing two simulations analytically. The result is the derivation of two equation models; one for each respective simulation. The derived models are analytically compared to determine: equivalence, consistency, linearity, similarity, and degree of overlap. This yields a unique analytical tool for comparing simulation versions or scenarios for VV&A. Methods as simple as regression can then be used to determine if accreditation is maintained on new simulations or models. The derived analytical functions can themselves be appropriately combined into an adaptive intelligent lookup table (LUT) equivalent model for real-time simulation purposes.

Multi-time scale unsupervised neural networks (MTSUNN) represent an established technique in pattern recognition for feature extraction and cluster analysis. From the nonlinear systems analysis perspective, they implement a very complex coupled multi-mode dynamics. This paper gives a comprehensive overview of several neural architectures of a combined activity and weights dynamics. The global asymptotic and exponential stability of the equilibrium points of these continuous-time recurrent systems whose weights are adapted based on unsupervised learning laws are mathematically analyzed. The derived architectures can lead to hybrid implementations in VLSI techniques.