Artificial Neural System for Object Classification and Orientation Estimation

摘要

The usefulness of self-organizing neural systems for the problems of objectrecognition and orientation estimation are discussed. Self-organizing neural systems, like unsupervised cluster algorithms from classical pattern recognition, are most useful when no predetermined labels are available to attach to input patterns which must be categorized by the system. However, this makes it necessary to assign a 'natural' category to the input stimuli from the environment. Only with some type of supervisory feedback will this natural category be associated with the proper label for the input pattern. Without this teaching input, the internal, self-organizing principles of the system must be used to assign categories. These assigned categories may or may not coincide with the unique labels of an external supervisory system.