A classifier for feature vectors whose prototypes are a function of multiple continuous parameters

A fast, compact continuous-parameter (CP) classifier, suitable for a 16-bit microprocessor, is developed for classes which consist of a prototype manifold which is a function of one or more continuous parameters. The classification method consists of approximating the manifold by a number of unit cells and assigning a test vector to the closest cell using a Euclidean distance measure. An experiment is described in which computer-generated magnetic dipole moments are used as feature vectors to classify a set of homogeneous ferrous spheroids. The CP classifier provides accurate estimates of the orientation angles of the test object with error equal to a small fraction of the design set increment (1° out of 15°)