Model the uncertainty in target recognition using possiblized bayes' theorem
Wei Mei, Ganlin Shan, Chunping Wang · 2012
The major source of uncertainty in target recognition consists of two parts. One is about feature extraction from observation data and another is about the rule definition between target type and feature. While the former can be naturally captured in the form of statistics, it is our opinion that the latter should be defined by using possibility since exact probability assignment is in general impossible. This paper addresses target recognition within the Bayesian framework while reinterpreting the likelihood of Bayes' theorem as a possibility. It leads to an open structure of feature database, which can exempt the reconstruction of feature database of the Bayesian classifier when new feature rules need to be included. An example of target recognition using attribute data from an electronic support measure (ESM) shows that the proposed method has competitive performance with the conventional Bayesian classifier.