Identification of unknown categories with probabilistic neural networks

The ability to identify correctly a pattern as an unknown as opposed to misclassifying it as a known category is a desired but often overlooked feature in all neural networks. The method described solves this problem by establishing a threshold on the probability density function (pdf) as determined by a risk strategy. Once sufficient numbers of samples of an unknown category have been collected, it can be added to the existing probabilistic neural network (PNN) classifier as a new category. This online real-time learning technique may be applied to many problems including voice recognition, optical character recognition, automatic target recognition, fault detection, and sonar processing