IJCNN 2015 Tutorial (T13) on Successful Applications of Neural Networks for Information Fusion
Where: Mangerton Suite, Killarney Convention Center in Killarney, Ireland
When: July 12, 2015. 13:45–15:45.
Information fusion is the problem of combining or fusing data from a variety of sources in order to provide a clearer understanding of the events being monitored and ease the associated decision making with respect to those events. Originally, these data sources were measured systems, such as radar and sonar systems, but now include complex and unmodeled data sources such as images, text, and human reports. The internal dynamics of the fusion process were once simple linear models that related the data spatially and temporally. Demands for increased accuracy, in near-real time and under complex scenarios, obviates the need to replace the simplistic models of the past. Beyond the accuracy issue, information fusion has grown from the basic applications of target tracking and object identification. Situational assessment and impact assessment have grown in application importance. Better understanding of the sensors’ behaviors has also become an increase area of concern.
For many years, neural networks have been used to model sensor variations, target motion, and identify objects using multiple features. In recent years, the use of intelligent systems, such as neural networks, have begun to provide the ability to model the intricate dynamic interactions of the various forms of the data and provide the robustness to handle ever changing temporal and spatial dynamics. As data sources continue to vary, new methods of mapping the data to a useful format are needed. Neural networks can provide the ability to generate these models. The decision process of information fusion requires models of relationships that must evolve and adapt in ways that overwhelm traditional modeling techniques and straightforward adaptive and rule-base methods. Without intelligent systems, including neural networks, the continued improvement of information fusion systems is limited.
Topics to be covered:
- Introduction to the JDL Fusion Model The Joint Directors’ Laboratory model of fusion decomposes information into six levels. These six levels look at sensors, objects, group behaviors, impact analysis, operator visualization, and refinement. From these six levels, we will focus on the first four and how neural networks have been used and where they can be applied.
- Level 0: Source preprocessing
Sensors provide measurements, but these signals need to be preprocessed into a useable form. Sensor errors need to be defined and adjusted. Example techniques include neural network for signal classification and sensor registration correction.
- Level 1: Object refinement
Object classification and kinematic generation are primary techniques in this level. Data is reported in different forms but must be combined. The complex relationships of different types and sources can be modeled using a neural network. Neural networks have also been used for generating target classification. Example Leve1 techniques include feature-data object classification and adaptive motion modeling using neural networks.
- Level 2: Situation Assessment
By examining the behavior of a single object or multiple objects along with their identification, an understanding is inferred for capability, interaction, and future actions. Neural networks have modeled the complex interactions of multiple objects to assess related objects and their capabilities. Example Level 2 techniques include object grouping and cyber security attack classification.
- Level 3: Impact Assessment
With coordinated objects and individual objects, along with an understanding of one’s own capabilities and weaknesses, the external-threats impact is understood. Neural networks provide a modeling technique for the complex reasoning used to infer the impact. Example Level 3 techniques include target threat assessment and cyber security threat assessment.
Organizers: Stephen Stubberud, Ph.D. and Kathleen A. Kramer, Ph.D.
Department of Electrical Engineering
University of San Diego
||Dr. Stephen Stubberud received his doctorate from the University of California Santa Barbara in Electrical Engineering. He has over twenty years of experience since then working in the fields of sensor data fusion and neural networks. He has worked on a number of fusion problems that include land, sea, and air tracking. His work has in multiplatform tracking systems gave him first-hand knowledge of sensor registration and techniques that exist to compensate for such errors. In the past ten years, Dr. Stubberud has researched techniques to compensate for registration and techniques to associate tracks from different platforms when they are registered. Dr. Stubberud has over 100 publications in the literature. He has worked for ORINCON, Raytheon Space and Airborne System, and Boeing. He is currently the senior research scientist at Oakridge Technology. Dr. Stubberud is a senior member of the IEEE.
||Prof. Kathleen A. Kramer is a Professor of Electrical Engineering at the University of San Diego. She received her doctorate in electrical engineering from the California Institute of Technology. She maintains an active research agenda and has over 90 publications in the areas of multi-sensor data fusion, intelligent systems, and neural and fuzzy systems. She has been a Member of Technical Staff at several companies, including ViaSat, Hewlett Packard, and Bell Communications Research. She is the 2015-16 Director-Elect for IEEE Region 6 (Western USA) and a member of the Board of Governors of the IEEE Aerospace Electronic Systems Society.
Register at IJCNN 2015
Disclaimer: The opinions expressed in this web page and the presentation slides are that of the organizer, not of the IJCNN conference or IEEE, or any other entity.
Updated: Sat Jun 6 18:38:00 PST 2015