Performance assessment of data fusion systems

Abstract

There are various architectures for data fusion. Given that the purpose of a data fusion system is to combine related data from multiple sources to provide enhanced information, one way of assessing the performance is to measure the enhancement or degradation in the information provided by the system and this is the point of view which we take in this paper. In order to achieve this goal, methods for measuring the information provided by the output of the system are required. The most commonly used method employs the relationship which exists between measures of information and measures of uncertainty. By convention, measures of uncertainty for a system can also be treated as measures of the potential information in the system. Therefore, the premise is that a decrease in uncertainty amounts to a decrease in the potential information, so that the information yielded from the system increases. The authors previously (1996) introduced concepts for a rule-based expert system which performs high level data fusion for human decision support. In this paper, we shall describe an updated version of this system, as well as some of the measures of information which we have devised for assessing the performance of the system. Finally, we shall discuss ways of combining these measures of information to gauge the enhancement or degradation in the information provided by the system.