Distributed detection theory represents an extension of conventional statistical decision theory that is applicable to the situation where a team of decision makers solve a hypothesis testing problem in a cooperative manner. The theory is particularly applicable to distributed multisensor data fusion problems with constrained data links. This topic wzas introduced here by discussing two problems in the area of decentralized decision-making. These problems represent two components of the overall distributed detection network using a parallel topology. The first problem focused on decision making by the individual team members while the second one considered the fusion of decisions made by the team members. Observations were assumed to be conditionally independent which simplified matters considerably. Optimization of the complete system in which both components are considered together is much more involved even under the conditional independence assumption. When this assumption is not valid, the problem has been shown to be NP-complete. Development of efficient computational algorithms for distributed detection problems is ongoing. Several algorithms such as the Gauss-Seidel cyclic coordinate descent algorithm have been employed successfully for system design. For a more complete understanding of these issues, the reader is referred to. The focus here was on the parallel fusion network topology and a Bayesian formulation of the problem. Distributed detection problems for other network topologies and under other formulations such as the Neyman-Pearson criterion have been discussed and solved. Open problems in this area include system design for the dependent observations case, and development of computational algorithms and paradigms to solve multiple hypothesis distributed decision problems for applications such as object recognition.
展开▼