Signal classification for distributed decision networks with uncertainties and unmodeled class distributions

An approach is presented for the classification of a signal in noise. The classifier presented forms the bottom level in a decision network. By determining an interval of doubt about classifications, it is possible to make decisions at higher levels with additional or conflicting evidence, without having biased the decision from a low level classification. The region of uncertainty is a function of the information, so that the quality of decisions from individual decision makers will vary with the input. The decisions are formed in a parallel network which has a similar connectivity to an artificial neural network.<>