Generalized probabilistic neural network based classifiers

Two new probabilistic neural-network-based maximum-likelihood classifiers are presented. These classifiers are based on Gram-Charlier series expansion with and without Parzen's windowing technique. The performance of the proposed classifiers are evaluated in terms of probability of target detection for a number of Gaussian and non-Gaussian noise source, and are compared with those of existing neural network classifiers, such as Bayesian and backpropagation classifiers. The new neural network classifiers performed better than existing classifiers in radar target detection. These classifiers are also applicable to many more practical situations than D.F. Specht's (1988) Bayesian classifier.<
>