Decentralized detection: Optimizing with Bayes’ or entropy-based Criterion?
Résumé
This contribution deals with binary detection networks optimization using an entropybased criterion. The optimization of a detection component consists in applying a variable threshold on the likelihood ratio, which depends on a posteriori probabilities. A gradient algorithm is proposed to find this threshold. The optimization results of the detection component using entropy and Bayes’ criteria are compared. Assuming conditional independence of the observations, the optimization of a complex parallel distributed detection network, constituted by several detection components, can be made using a person-by-person technique, whatever the considered local criterion (Bayes’ or entropy-based criterion): a compromise between a good error probability and a good ROC curve is obtained.