A Pattern Recognition Model for on-Line Curve Fitting: An Application of Threshold Theory

摘要

The report describes the development of an adaptive pattern classifier in application to curve fitting problems. The research develops additional theory for general pattern classifiers; i.e., one-step correction under various training methods, and development of a modified Euclidean metric. Two adaptive pattern classifiers have been designed for solutions of practical scientific problems. Heuristics for learning are coupled to threshold logical theory. In application the curve fitting model classifies realistic patterns at 200 to 300 patterns per minute on the Burroughs B5500 with average error of 3.5%. The method does not require the 'good' initial curve coefficient guesses required by other approaches.